A short week that makes joining a team by email work and puts pinned widgets first on Home.
Email sign-in for new teammates
People invited without a Google or Microsoft account can now create their account by signing in with email.
Improvements
Pinned widgets now sit at the top of Home, above the getting-started checklist, once your organization has pins.
Fixes
Fixed sign-in links sending people to the generic dashboard instead of the page they started from.
Week 41
Signup now starts from what you want to do, the sidebar chat follows you across pages, and integrations gain new actions and triggers.
Signup that starts from your goal
New users now pick what they want first, like an agent in Slack or Teams, a Claude artifact, a company brain or a sheet, and the signup flow branches from there.
Chat that follows you across pages
The sidebar chat now stays open as you move between pages, shows every tool step, and lists your integrations in the / menu.
Granola actions and triggers
Agents and workflows can now run Granola actions and react to new Granola events with polling triggers.
Apps know what you sell
Apps can now read your offering from the Company Knowledge document pinned in Settings → Brand, so their AI features describe your business without a separate settings field.
Chat that survives updates
Chat turns now carry on through platform updates, and a Retry button appears when a message got no reply.
Improvements
Stop now keeps the partial reply instead of discarding it.
A group of tool steps in chat turns red only when its last step failed.
The agent you pick in the Triggers dialog is now the one that receives the events.
Rebuilding or updating an app now also refreshes its agent instructions and what it offers to other apps.
GitHub repos are now linked from Settings → GitHub only, instead of from popovers in each navbar.
Every agent's Chrome now includes TaskWindow, our browser extension for agents.
An agent now stays awake instead of resting when waking would remove a tool it relies on.
Only admins can now delete Company Knowledge documents.
Integration logos now show the right brand.
Integrations
ElevenLabs: agents and workflows can now run ElevenLabs actions.
GitHub: agents and workflows can now run GitHub actions with a personal access token.
Ahrefs: agents and workflows can now run 16 Ahrefs actions.
Autotask: agents and workflows can now run 16 Autotask actions.
Gmail: an organization can now connect a second mailbox, and each connection says which mailbox it is.
For developers
@clawnify/knowledge gives apps offering(env), which returns the organization's pinned offering document or null.
@clawnify/connections now calls the provider's own API through the proxy first, with rawRequest and a new rawFile for files.
clawnify_app_api_write now says whether an edit's save was confirmed, not only when it failed.
Video rendering can now run in the background with /video/render and async: true, in SDK 0.4.0.
Clawnify Services /video/edit adds transitions between clips, fades on clips and overlays, volume keyframes on audio and colour adjustment.
Clawnify Services /video/capture now records what the page does after each click as a short video.
Clawnify Services adds /media/transcode to re-encode footage the video host refuses for its bitrate.
Importing footage and analysing video no longer count against the plan's monthly render allowance.
Fixes
Fixed the sidebar showing a stale agent status after going back to Home.
Fixed a finished re-encode being imported more than once.
Week 40
Agents now rest when idle and wake when needed, and apps gain live presence, home widgets and several accounts per integration.
Resting agents that wake on demand
An agent that is off now shows as "Dreaming" and wakes when you open it or when a scheduled task is due, and email, webhooks and tasks sent while it rests are held until it can take them.
See who is in which app
Teammates' avatars now appear beside Apps in the sidebar and in each app's top bar, so you can see who is working where.
Home widgets from your apps
Owners can now add widgets built from app data to the dashboard home from the + button, pin them for the whole organization, and arrange them in a packed grid.
Several accounts per integration
You can now connect, reconnect and disconnect more than one account for the same service, and each app can choose which account it uses.
Sidebar chat with history and commands
The sidebar chat now keeps a history of past chats with generated titles, offers / commands, and runs each chat on its own so two browser tabs can work on different chats at once.
Apps and chat that drive each other
An app can now open the sidebar chat with a drafted prompt, and the chat can open any page of an app.
You can now put your apps in the order you want in the sidebar.
Owners and admins can upload an organization logo on the Members page, and it shows in the sidebar and browser tab.
Agents now know about their upcoming scheduled work.
Agents can now read long app fields in parts and edit them in place.
The sidebar chat can now send an image from the conversation into an app.
Nightly backups now cover the agent's whole home folder and its scheduled jobs, so nothing it saved is lost when it wakes.
New agents now run OpenClaw 2026.9.4.
Agency staff can now work inside the client workspaces their agency manages.
Company Knowledge documents can now be created without a description.
Website video and audio now stream in byte ranges, so they play on iPhone.
Integrations
Google Drive: agents now find files in shared drives when they list files.
Granola: apps can now reach Granola's API through Clawnify connections.
Models
Pro and Lite in the US now run on newer models.
For developers
clawnify.json can declare app.provides so other apps can find a compatible sibling, like a CRM that provides contacts.
GET /v1/apps/directory lets a deployed app list the live apps in its organization.
Apps can now draw their own sidebar and chat when they run outside the Clawnify dashboard.
@clawnify/services is now the one client for every Clawnify service.
@clawnify/services 0.2.0 adds /video/capture, which walks a live page step by step and screenshots each step for product demos.
Clawnify Services adds /markdown/render, which reads a live page as markdown, including JavaScript-built sites.
Clawnify Services /video/edit can now outline text with a stroke, fade audio in and out, and anchor a crop so an off-centre speaker stays in frame.
The CLI now names --discard-draft when a deploy is refused because of a pending draft.
Fixes
Fixed signing out from the sidebar menu doing nothing.
Fixed a deploy publishing a pending app draft you had not approved.
Fixed Stop not reaching a chat turn that resumed on its own.
Fixed Make admin and Make member not actually changing a member's role.
Fixed reconnecting an integration losing the triggers it was wired to.
Fixed Granola requests from apps answering that no connection exists.
Fixed the dashboard breaking in a tab left open across an update.
Fixed an old build error staying on screen after a GitHub or CLI deploy succeeded.
Fixed the dashboard offering Publish for an app rebuild that was already live.
Fixed a template upgrade that adds files failing to build.
Fixed draft websites dropping files of 5 MB or more.
Fixed video renders failing when they use uploaded images or clips.
Week 39
Microsoft Teams arrives as a channel, and owners get more control over who can reach an agent and what it may do there.
Microsoft Teams channel
Agents can now take messages in Microsoft Teams, connected from the agent's Channels dialog with your own Teams bot and a ready-made app package to upload.
Choose who can message your agent
Owners can now approve who may message an agent directly from the Channels dialog, and on Slack pick the members who can reach it before they ask.
Approve your agent's actions in chat apps
Owners can now approve the actions an agent wants to take in Slack, Telegram and Teams before they go out.
Nightly backups of every agent
Each agent's state is now backed up off its machine every night, so it can be restored if something goes wrong.
AI disclosure on outgoing email
Agent emails to anyone other than the owner now end with "Composed by <Name>, an AI agent.", as the EU AI Act requires.
Clawnify plugin for Claude
Clawnify's skills and MCP connector now install together as one plugin in claude.ai, Cowork and Claude Code.
Apps list the keys and settings they need, and a key you save in Settings reaches them right away with no redeploy.
New agents start from a prebuilt image and are ready to chat in about 100 seconds.
Agents no longer end a turn on a promise of work they have not done.
Agents reading a large app response get an outline of it instead of a list that was silently cut off.
The usage page now breaks spend down by source and by hour, and labels video analysis.
Billing is now owner-only, and an org on a live plan opens on its plan summary with the plan grid behind Change plan.
A live website now shows Publish after design-tab, chat or favicon edits that need no build.
Changing the site icon no longer rebuilds the website.
A root custom domain now redirects to www when both are connected, and connecting www also registers the root.
The website and app status line now says where a build is, from syncing a GitHub commit to building and publishing.
Integrations
Slack: an @mention in a channel now starts its own thread with the agent.
Slack: the panel now says when a token is wrong instead of showing Connected.
Google Workspace: new connections no longer show an unverified-app warning during sign-in.
Models
Agents now recover faster when a model goes silent.
When a provider errors partway through a reply, the agent now carries on.
For developers
CLI 0.17.0 refuses a deploy that would revert newer changes in the synced GitHub repo, unless you pass --force.
The API refuses a website deploy that would replace the org's existing site.
@clawnify/routes 0.3.0 and @clawnify/app 0.2.2 add paged list route helpers, and CLI 0.19.0 templates use them.
Large results from the Clawnify MCP connector now come back in pages instead of failing, and company_knowledge_get returns very long documents in parts.
Clawnify Services adds /video/analyze to watch and listen to a whole video in the EU.
Clawnify Services /video/edit now runs in the background and can cut from long uploaded media.
Clawnify Services adds /media for resumable uploads of up to 30 GB, imports from a link and AI captions.
Fixes
Fixed signing out disconnecting your Claude and ChatGPT connectors.
Fixed favicon upload failing with "Expected multipart/form-data".
Fixed CLI deploys deleting a website's live content and uploaded images.
Fixed synced website builds stopping partway and leaving the site stuck on building.
Fixed removing a member or cancelling an invite reporting success without doing it.
Fixed a paid workspace being locked out after an ended subscription overwrote its live one.
Fixed agents being restarted mid-task by automatic system updates.
Fixed agents unable to convert documents on new machines.
Week 38
A Usage page that shows where credits went, agents that can search their own inbox, and every scheduled flow on one Schedule list.
See where your credits went
Owners can now open Usage from the account menu to see credits used, daily spend, and spend by agent, by plan tier and by hour of day.
Agents can search their own inbox
Agents can now search and read their own email threads from chat, so a question like "did Ann reply?" gets an answer, and plain-text mail is kept for 12 months.
Every schedule on one list
Flows an agent schedules now appear on the Flows and Schedule pages, and you can pick "Run a flow" when you create a scheduled task.
You can now delete a single email, or all mail with one person, from an agent's inbox in the dashboard.
Workspaces without an active subscription now see a billing page in place of the dashboard, and their apps and agents pause until they resubscribe.
Template apps now have shared demo workspaces, with sample data and file uploads, that you can try before deploying.
For developers
Apps: integrations can now call an app at its own hostname or custom domain with an Authorization: Bearer access token (cpat_…) or org token (clw_…).
@clawnify/agents: new client for the org-scoped schedule API, with request receipts and pagination.
@clawnify/openclaw-email 0.2.0 adds search_email and read_email_thread tools.
Fixes
Fixed agents replying to an email without knowing which message they were answering.
Fixed the send email step failing in flows.
Fixed lapsed subscribers seeing "No active subscription" instead of a way to reactivate.
Fixed the billing page opening on the smallest credit tier instead of your current one, which could downgrade a plan on confirm.
Fixed apps in older workspaces missing their built-in AI key.
Fixed dashboard links losing encoded characters in their query.
Week 37
One task board for people and agents, apps that join the dashboard sidebar, and Company Knowledge that says whether a document still holds.
Tasks for people and agents
Teams can now keep one Tasks board in the dashboard where each task goes to a person or an agent, and an agent task runs on save and shows its result, like a summary or the reason it failed.
App sections in the dashboard sidebar
Apps can now put their own sections, like Contacts or Deals, into the dashboard sidebar through <AppNav> from @clawnify/app/client, instead of drawing a second sidebar inside the app.
Owners and admins can now mark a document verified for 90 days or flag it outdated, and agents rank verified documents higher and never see outdated ones.
A cleaner look for generated apps
Apps the builder makes now start from a white, ring-edged design with a generated colour palette, checked by a design lint on every build.
Faster repeat deploys from a template
Deploying a template at a commit that was already built now reuses that build, so later installs skip the build step.
Improvements
Tasks: type @ in a task title to tag a person or an agent.
Tasks: an agent stands down when its task is marked complete.
Tasks: the board now works on phones and tablets.
Tasks: the task composer now wraps and grows as you type instead of scrolling.
Tasks: the board sits right after Chat in the sidebar.
The workspace chat can now create, update and delete tasks.
Website forms can now send visitors to a page of your choice after they submit.
A GitHub push to a website now applies only the files it changed, so it no longer reverts edits you just published from the dashboard.
App sign-ins now renew while you use an app, and the app says so when access lapses instead of failing quietly.
Apps deployed from a template or the Deploy button now publish right away, and only edits made from the chat stay as drafts.
Dialogs and popovers in the dashboard now use the frosted edge from the homepage.
The dashboard sidebar now uses one icon set, with outline app icons and tiled section icons.
The sidebar peek now floats as a card and closes when you move away.
For developers
MCP: Company Knowledge reads now return Markdown by default, with JSON on request.
MCP: company_knowledge_write now makes revision-safe text edits and can publish in the same call with publish: true.
MCP: new company_knowledge_review, company_knowledge_archive and company_knowledge_restore tools.
MCP: every Company Knowledge tool now carries read-only or destructive annotations, so clients can set permissions per tool.
MCP: new task tools let Claude, ChatGPT and Cursor read and manage the Tasks board.
API: /v1/tasks lets apps read and write the org's tasks.
API: POST /v1/knowledge/:id/review sets a document's review state.
API: POST /v1/bundles/install deploys every app in a bundle repo in one call.
@clawnify/app 0.2.1 publishes the Phosphor-based @clawnify/app/client entry.
MCP: flow-run results are now capped in size, and a rate-limit error names which limit fired.
Fixes
Fixed apps that answer with a redirect showing their home screen instead of redirecting.
Fixed the prompt from a Deploy button or an integration hand-off being lost on the way to the agent chat.
Fixed a Company Knowledge publish reporting a failure when it had actually published.
Fixed editing a published Company Knowledge document over MCP changing the live version in place.
Fixed the website editor drawing native radio buttons over the design after an edit.
Fixed the website editor not finding the fields of some blocks.
Fixed long flow runs crashing the MCP session when an agent checked on them.
Fixed Home not highlighting in the sidebar on the dashboard.
Fixed Tasks showing up twice in the sidebar.
Week 36
EU data residency for model calls, one place to set what agents may do on their own, and more integrations connected directly.
Keep model calls in the EU
Owners on paid plans can now pin every model call the platform makes for their org to EU providers from Settings → Data region, with a US option too.
Agent permissions in one place
Owners can now see and edit what agents may do on their own from Settings → Permissions, with each rule set to allow, ask or deny.
Template apps that keep up with their template
Apps deployed from a template now notice when the template has changed, and the agent can carry those upstream changes into your app instead of you redeploying from scratch.
The Deploy with Clawnify button now asks which workspace should receive the app when you belong to more than one.
Editing a published Company Knowledge document now creates a draft revision, and the live version stays unchanged until it is published.
The dashboard home now shows Concepts cards for company brain, integrations, flows, schedule and website, each linking to where you use it.
The sidebar toggle now sits next to the workspace switcher, and hovering the icon rail opens the full sidebar as an overlay.
Apps in the dashboard can now use the microphone, so an app can place browser calls with Twilio.
Integrations whose connection check can never pass are no longer offered.
Integrations
Apollo: now connects directly with your API key, verified when you connect.
Perplexity: now connects directly with your API key, verified when you connect.
Ahrefs: now connects directly with your API key, verified when you connect.
Supabase: now connects directly with your key, verified when you connect.
DataForSEO: now connects directly with your login and password, verified when you connect.
SerpApi: now connects directly with your API key.
Facebook: the integration card now covers publishing to Facebook Pages.
WhatsApp: a reaction emoji now shows in the Inbox instead of an empty bubble.
Models
With the EU data region on, Lite keeps working as usual, served from EU providers.
Platform tasks and media now also stay in your data region.
For developers
MCP: the app lifecycle tools are back on the connector, behind your org's permission rules.
MCP: rate limits now apply per caller under an org-wide ceiling, so one automation in a retry loop no longer locks out everyone else.
@clawnify/connections: credentials() reads every field of a multi-field credential, such as a Bluesky identifier and password.
@clawnify/connections: stageFile posts media to a provider through the broker.
@clawnify/connections: integrations with a hyphen in their id, like whatsapp-business, now resolve their token from the environment.
Apps: WhatsApp Business account settings are now available to apps that connect WhatsApp.
Apps: an app that declares TWILIO_ACCOUNT_SID and TWILIO_AUTH_TOKEN in clawnify.json now receives both.
Services: /screenshot/render now accepts a live URL as well as HTML.
Fixes
Fixed a provider's content-type header making an integration call return an empty result.
Fixed residential browsing not following a plan change until the owner toggled it again.
Fixed adding a permission rule failing.
Fixed a message you just sent disappearing from the chat before its turn had run.
Week 35
Your workspace on WhatsApp, a chat that edits your website itself, and residential browsing for agents.
Clawnify on WhatsApp
Members can now link their own WhatsApp number from Settings → WhatsApp and text Clawnify's number to reach the same workspace chat they use in the dashboard.
Approve or reject from WhatsApp
Actions that need your sign-off, like publishing, now pause and wait for you, and you can approve or reject them with a button in WhatsApp or in the dashboard.
Residential browsing for agents
Owners on Team plans and above can now let their agents browse through a dedicated residential IP per agent, so fewer sites treat them as bots.
Your brand brief in every build
Owners can now pin one published Company Knowledge document as the brand brief, and every app and website the workspace builds follows it.
Google Business Profile integration
You can now connect Google Business Profile so agents and apps can work with your business listing.
The chat edits your website itself
The workspace chat now edits your draft site directly, seeding the source, changing files and deploying the draft, instead of handing a text instruction to a separate build.
Improvements
The workspace chat now takes a screenshot of the section it changed on your draft site and checks it before reporting back.
Images you drop into the chat can now become site images, served from your website's /assets on both draft and live.
Images uploaded in the dashboard and images added from the chat now share one asset store per website.
Theme edits from the chat now apply instantly instead of starting a build.
The workspace chat can now act on your org's connected integrations directly, the same way agents do.
Each linked WhatsApp number now gets its own chat thread.
You can now link more than one WhatsApp number to your account.
Photos you send on WhatsApp now reach the workspace chat.
A new "Get set up" checklist on the dashboard walks through building company knowledge, connecting Claude and ChatGPT, inviting your team and finding a first use case.
Company Knowledge now has a search bar.
Company Knowledge now works properly on phones.
The website editor now works on phones, with a tab strip and full-page panels over the preview.
The website editor now offers an image picker for gallery and media fields.
Draft website builds now reach the preview faster, without a separate deploy step.
Integrations
Google Business Profile: new integration for managing your business listing.
Google: the Google connection now includes Drive and Calendar access.
For developers
Apps: public routes now serve stale content while refreshing in the background, honouring stale-while-revalidate instead of stalling on every cache expiry.
Apps: a public response to a request carrying a cookie or an Authorization header is no longer stored in the edge cache.
Apps: a pending migration that only creates tables, indexes or columns that already exist is now recorded and skipped instead of failing the publish.
Apps: /api paths listed in api.server_routes in clawnify.json are now dropped as redundant instead of failing the deploy.
Fixes
Fixed draft websites getting an error on every request to one of the org's apps, which left app-fed sections empty in the preview.
Fixed the chat crashing while replies were loading.
Fixed a message you just sent disappearing from the chat before the reply arrived.
Fixed the chat saying an edit was live on the draft while the draft was still showing an older build.
Fixed the workspace chat reading the published site instead of the draft when answering about your website.
Fixed chat edits to nested website fields, like one item in a list, not applying or touching the wrong field.
Fixed entrance animations on draft sites reloading the page while you edited.
Fixed the website editor selecting elements on the wrong page.
Fixed image edits in the website editor not applying.
Fixed the website editor sometimes missing a change until you reloaded.
Fixed the chat panel, sidebar and model picker reverting to an old state after navigating back to the dashboard.
Fixed paid plans showing a trial state, and losing automatic top-up, after adding an agent.
Fixed a canceled subscription still showing as active in the dashboard.
Fixed an app's first migration leaving a wrong saved schema behind for later builds.
Fixed the Company Knowledge MCP write tool telling models not to call it.
Week 34
One chat docked across the dashboard, scheduled tasks, automatic credit top-up, client workspaces for agencies, and card-free deploys from coding agents.
One chat across the dashboard
The chat now stays docked as you move around the dashboard, takes images, has a Stop button and a model picker, and can create apps and edit agent instructions and workflows for you.
Scheduled tasks
A new Schedule page in each agent's sidebar lets you create, edit, pause, run and review recurring tasks, set with presets like daily, weekdays or monthly.
Automatic credit top-up
Owners can now have credits refill automatically and get an email when credits run low, at 10% of the plan's allowance by default.
Client workspaces for agencies
Agencies can now create a workspace for each client from the workspace menu, where their own team gets access automatically and billing runs through the agency.
You can now connect n8n, Make, CI jobs and other MCP clients that cannot sign in with OAuth using scoped, revocable personal tokens from Settings → Access tokens.
The website editor now uses the same docked chat, with Design next to Layers on the left.
Creating a website from the dashboard now opens the chat with a draft prompt ready.
The dashboard chat can now apply schema migrations when it changes an app.
The dashboard chat now answers questions about an app from its brief and API index before reading its source.
Chat now sits at the top of the sidebar, right after Home.
The credits dialog now shows buying credits and automatic top-up side by side.
Template apps from verified repositories now carry a trust badge and always deploy the exact version that was checked.
Website forms can now accept PDF attachments.
Websites now ship a working favicon, and you can set your own.
People invited into a workspace no longer land in the sign-up flow when it has no agents yet.
Integrations you connect with your own key now use that key directly.
Models
Google keys you bring now use Gemini 3.7 Flash as the fast model.
For developers
CLI 0.16.0 adds anonymous deploys that return a claim link.
The CLI now keeps the drizzle config inside .clawnify.
Personal access tokens (cpat_…) work as a Bearer header on the MCP connector URL and expire by default.
Code execution over MCP is now off until an owner turns it on for the workspace.
An app's /api/* 404 now lists the endpoints the app really serves.
MCP knowledge tools now tell token callers where a person publishes a proposal.
The docs now offer hosted setup instructions you can hand to a coding agent in one line.
Fixes
Fixed chat replies vanishing at the end of a turn.
Fixed the chat flashing an error for a turn that had succeeded.
Fixed the chat losing the error line when a turn stopped mid-work.
Fixed the chat getting stuck in a crash loop after a failed turn.
Fixed the Thinking indicator disappearing during pauses in a live reply.
Fixed the chat stopping silently when it reached its step limit.
Fixed long links overflowing chat bubbles in the sidebar.
Fixed the sidebar's empty states when it is collapsed to icons.
Fixed changing the main agent's avatar not updating the image.
Fixed app calls to sibling apps and Clawnify Services failing intermittently.
Fixed connection errors in apps showing as "[object Object]" instead of the provider's message.
Week 33
GitHub sync for the whole workspace, clearer custom domain setup, workflow tools for connected assistants, and video editing for apps.
GitHub sync for the whole workspace
You can now manage GitHub sync for every agent, app and website from one page in Settings, and syncing an agent brings its sub-agents with it, each in its own folder.
Manage workflows from Claude, ChatGPT and Cursor
Six new MCP tools let a connected assistant list, read, edit, publish and run ClawFlow workflows and inspect their runs, with no terminal needed.
Apps can now call a connected service's own API when no ready-made action covers the job, like editing a WhatsApp message template.
Video editing for apps
Apps can now cut, trim and overlay footage from an edit list through Clawnify Services, including playing a set number of seconds from a trim point.
Clearer custom domain setup
DNS instructions now stay on screen until every hostname is connected, each hostname needs two records instead of three, and a domain is no longer shown as live before it can be reached.
Improvements
A website with a working www address now treats it as the site's real address.
GitHub sync now names an agent's repository folder after the agent instead of "main".
GitHub sync now offers the right flow when the GitHub account is installed somewhere else, and opens the right popover after an install.
Connecting GitHub now activates one GitHub account at a time instead of every account you own.
New workspace repositories now include a clawnify.json, so tools run inside a clone target the right workspace.
Uploaded website images are now cached at the edge, so they load faster.
Integrations
WhatsApp: contact cards that people share with your agent now come through.
GitHub: links to settings now go to GitHub's own settings page.
Models
Agent workflow runs no longer get derailed by quality checks meant for chat replies.
For developers
CLI 0.14.0 refuses an org you cannot use instead of quietly answering as a different one.
CLI 0.15.0 removes the machine-wide active org, so each project resolves its own workspace.
clawnify deploy now stops when an app needs a database but has no schema to apply, instead of shipping an app whose queries fail.
clawnify deploy now retries uploads that drop on a network error.
clawnify init apps now include db:generate and typecheck scripts and light and dark design tokens.
Apps can declare their own Worker-served routes in clawnify.json under api.server_routes.
Clawnify Services adds /video/edit for edit-list-based footage editing and /files/presign for staging files.
The build-a-clawnify-website skill is now public.
Fixes
Fixed tool activity from earlier turns piling up at the bottom of the chat.
Fixed DNS records being hidden before every hostname was connected.
Fixed a custom domain being shown as live while it could not be reached.
Fixed uploaded website images disappearing after a GitHub push.
Fixed a failed website publish that could not be retried.
Fixed a brief connection drop failing a website publish.
Fixed a website being deployed as if it were an app.
Fixed pnpm dev not starting a website locally.
Fixed a website loading the wrong webfonts for its design.
Fixed italic accents in serif fonts rendering as plain text.
Fixed website builds shipping leftover files from an earlier build.
Fixed app previews stuck on "Your latest change hasn't reached the preview yet".
Fixed clawnify deploy --from ignoring the folder and the workspace it was run in.
Fixed apps deployed from a repository missing their address in the dashboard.
Fixed credits dropping to zero for about an hour when a plan renewed.
Fixed the credit gauge measuring use against every credit ever granted instead of the plan allowance.
Fixed a new credit grant counting old usage against it again.
Week 32
Live workflow previews in chat, website analytics, inbound WhatsApp, and apps that call each other.
Live workflow preview in chat
When the chat edits a workflow while you watch, a split view now opens and shows each new step and connection being added with animated diffs.
You can now start a website's first draft yourself from the editor.
You can now open an app at a specific page from the dashboard.
The deploy page now previews a template's image and its button reads "Add to your workspace".
Public pages in apps are now cached close to visitors, so they load faster.
An empty Company Knowledge base now offers to import from Claude, ChatGPT or Obsidian.
Codex now appears in the list of assistants you can connect to Company Knowledge.
The Company Knowledge editor now shows clearly when a draft has been saved.
The agent listing now includes each agent's workflows and connected integrations.
GitHub sync now refuses a repository link that overlaps another one and defaults to the right folder in a monorepo.
Integrations
WhatsApp: Cloud API webhooks now connect from the integration settings and use the signing secret Meta issues.
For developers
CLI 0.12.0 adds clawnify flows create, edit and publish.
CLI 0.12.1 ships bundled skills that document clawnify flows.
clawnify ai-files install now writes rules that match the project kind, so websites and agents no longer get app rules.
The CLI and MCP tools now list agents and their sub-agents only.
clawnify logs now shows the last error of a failed app build.
Fixes
Fixed a GitHub push quietly reverting an agent's instructions.
Fixed agent files being deleted when GitHub sync could not read them.
Fixed the workflow change preview vanishing from the chat.
Fixed messages with attachments appearing twice in the chat.
Fixed website pages missing link previews when shared on social media.
Fixed websites served over http not redirecting to https.
Fixed website builds getting stuck when a build message was lost.
Fixed being unable to retry a website publish after it failed.
Fixed invite links that added a nameless member and left the invitee without access.
Week 31
Apps can hand work to your agent, websites take custom code, and Company Knowledge edits get a proper review.
Apps can hand work to your agent
Apps can now pass a task to your workspace's agent when the job needs judgment, a real browser or minutes of runtime, and the agent reports back through the app's own API.
You can now add analytics, Google Tag Manager or any other snippet to your published website from a Custom code dialog, placed in the head or body and limited to the paths you choose.
Review knowledge changes before they land
Proposed Company Knowledge edits now arrive as approval cards with the proposer's summary and word counts, and open into a diff that owners and admins approve or reject.
Website edits show in about a second
Text and theme changes the website assistant makes now appear in the draft in about a second, without waiting for a rebuild.
File uploaded documents in one click
Unfiled uploads in Company Knowledge now show as cards where you set a folder and tags inline, or select several and let Auto-file sort them into your existing folders.
Improvements
Company Knowledge drafts now save as you type.
A Company Knowledge document with pending revisions now shows an "N pending revisions" pill instead of appearing twice in the folder tree.
The review queue now includes revisions proposed from Claude and other connected assistants, not only agent drafts.
Every website page now has a Markdown version at its own /llms.txt path for AI assistants, which you can override page by page.
Websites can now add their dynamic pages to the sitemap.
You can now drag to reorder, remove and add website sections from the Layers panel in the editor.
The website editor now updates its status, canvas and chat live instead of checking every few seconds.
New websites ship with a default favicon instead of the browser's globe.
Website build output is no longer synced into your workspace's GitHub repository.
Newly built apps get understated ink links, a header line that runs straight across the screen and edge-to-edge tables.
The app builder now follows the full design brief on every build.
The dashboard sidebar now has a Home option.
Agents now start faster thanks to leaner built-in instructions.
Models
Replies that fail a quality check are now retried instead of sent as is.
For developers
/v1/agents/tasks lets an app send its agent an instruction and an optional payload.
/v1/agents/tasks accepts an idempotency_key, so a retried request runs the task once.
Clawnify Services adds /docx/redline, which turns a contract and a set of anchored changes into a tracked-changes Word document.
@clawnify/queue is now published to npm, so apps can use the queue service without hand-written HTTP calls.
company_knowledge_propose takes a change summary that reviewers see on the approval card.
Agents now discover an app's API from its instructions and source, and the unreliable get_app_openapi tool is gone.
Fixes
Fixed plan and credit tier changes from billing not taking effect.
Fixed credit top-ups missing from the invoice history in the billing portal.
Fixed public app pages returning an access error after a new agent was added.
Fixed website forms failing their spam check on custom domains.
Fixed a brief hiccup during a website build failing the whole site.
Fixed website chat edits on image-heavy sites ending with no reply.
Fixed the website chat staying on a pending reply until you reloaded the page.
Fixed website chat edits waiting forever when the build never reported back.
Fixed website source files being overwritten when two edits landed at the same time.
Fixed not knowing which GitHub account a new app repository would be created in.
Fixed clawnify deploy leaving untracked files out of the upload in large repositories.
Fixed clawnify deploy --from creating a new app on every run instead of updating the existing one.
Fixed clawnify --version always reporting 0.1.0.
Fixed an app's /llms.txt not reaching the app itself.
Week 30
Custom domains for apps, a Design inspector in the website editor, and GitHub sync that coding agents can set up themselves.
Custom domains for apps
Apps with public pages can now run on your own domain, and Clawnify detects your DNS provider among 50 and links you straight to its DNS settings.
The website editor now has a Design panel for a block's content, layout, style and text, with click-to-upload for images.
GitHub sync without the dashboard
Coding agents like Claude Code can now turn on GitHub sync with clawnify workspace connect and get a repo seeded with an AGENTS.md that explains how editing works.
Deployed apps can now show the signed-in person's name and profile picture through user(c) from @clawnify/app, on the live app, the preview and the API.
Fixed the GitHub account list staying empty after disconnecting an account.
Fixed the website editor overflowing the screen and long words spilling out of chat bubbles.
Fixed app icons not loading in deployed apps.
Fixed call_app_api dropping the query string when an agent called an app.
Fixed update_app asking agents to clarify when the instruction was sent under the wrong field name.
Week 29
A Notion-style editor and folder uploads for Company Knowledge, live website previews, and agents you can inspect and edit from Claude, ChatGPT or the CLI.
Notion-style knowledge editor
Company Knowledge and project docs now open in a rich editor with a properties panel and a link you can share to any document, while the stored text stays plain markdown that agents read.
Upload a whole folder to Company Knowledge
Teams can now drop a nested folder into Company Knowledge and have it organized into folders, tags and shared playbooks that cite their source files.
Live website preview
Edits to a website now appear in the preview in place, without the page flashing white or losing your scroll position.
WhatsApp can message any number
Owners can now let an agent start WhatsApp conversations with any number from the channel settings, while the allowlist still decides who gets a reply.
Manage agents from Claude and ChatGPT
The Clawnify connector can now list an agent's sessions, read transcripts, show its skills and edit its instructions from Claude, ChatGPT or Claude Desktop.
The integrations settings page now loads in about 10 ms instead of 3 seconds.
New agents now start with much less background loaded, so the first reply comes sooner.
Agents now always ask for approval before creating, updating, publishing or deleting an app or website.
Long lists returned by integrations, like a Drive folder, now come back as a compact table so no rows are cut off.
Website edits now rebuild a few seconds faster when the site's packages have not changed.
Agents now repair their disk and come back online on their own after an unexpected shutdown.
Owners can now top up the exact number of credits they need instead of picking a fixed pack.
The Change plan page now opens on your actual billing period and marks the current plan for that period.
The pricing page's agency view is now called For Service Companies, with a card linking between the two views.
Scrollbars across the dashboard are now subtler.
Models
Clawnify Lite now thinks before answering again.
For developers
clawnify sessions lists and shows the history of sessions across every agent in your org.
clawnify agents list, agents pull and agents push read and update an agent's instructions, skills and flows.
clawnify agents create adds a sub-agent from the terminal.
clawnify agents skills lists an agent's skills, and agents grant-skill shares one with an agent or with all of them.
clawnify connections connect <id> connects an integration after checking the id against the catalog.
clawnify env set and env unset manage your org's environment variables.
New fix-a-clawnify-agent skill teaches coding agents to diagnose and fix an agent.
New MCP tools clawnify_list_agent_sessions, clawnify_get_session_transcript and clawnify_list_agent_skills let connected clients inspect agents.
New MCP tools clawnify_get_agent_instructions and clawnify_set_agent_instructions read and replace an agent's instructions.
The clawnify_list_specialists MCP tool is now clawnify_list_agents, with sub-agents nested under each agent.
Agents now find integration actions with one clawnify_find_actions tool.
The get_app_openapi agent tool is removed, and agents learn an app's API from get_app_source instead.
Fixes
Fixed monthly credit grants not being applied to the balance.
Fixed credit top-ups being charged in the buyer's currency instead of the organization's.
Fixed queued messages being sent while the agent was still replying.
Fixed a finished website edit not showing its steps until the next message.
Fixed images shared with the website assistant being rejected.
Fixed Split View on mobile putting the chat above the artifact or browser.
Fixed sub-agents finding nothing when searching their own project's knowledge.
Fixed automatic compaction firing at the wrong point in long conversations.
Fixed ServiceOS plans being refused managed services like PDF rendering.
Fixed the clawnify_list_apps MCP tool failing to return apps.
Week 28
The agent's browser moves into view, sub-agents get roles and skills, and chat shows how full a conversation is.
Browser in the agent view
Owners can now open the agent's live browser from the sidebar next to the chat and limit which sites it may visit, while each agent keeps to its own tabs.
Context gauge in the chat box
The composer now shows how full the conversation is, and the transcript marks where older history was summarized.
Agent files show up in chat
Files an agent creates, like a PDF report or a downloaded image, now preview directly in the conversation instead of as a broken link.
Sub-agents with roles and skills
Sub-agents can now have a role, an image avatar and skills shared from the main agent, and you can switch between agents from the header or delete one you no longer need.
Website assistant takes images and keeps chats
The website assistant now accepts up to 4 images as visual reference and keeps named chat sessions you can come back to.
Agents read spreadsheets and Office files
Agents can now read Excel, Word, PowerPoint, CSV and PDF files as clean text and tables, so a spreadsheet sent in chat is no longer a dead end.
Improvements
Agents can now run heavy Python jobs on their own, so a big data task cannot slow down or crash the agent.
Invite links now work whichever email address the recipient signs in with, are single-use and expire after 7 days.
Agents now use your own timezone, captured when the agent is created, for messages and schedules.
New sub-agents now close the create dialog at once and show an Onboarding badge until they are ready.
The integrations page now shows Connected integrations above the ones still available.
Company Knowledge has a new Map view that lays out documents by folder with their links.
Markdown files uploaded to Company Knowledge now become draft documents, with a Needs filing queue for untagged ones.
Owners can now raise or lower monthly credits within their current plan, with a note on when the change takes effect.
Each credit tier on the pricing and billing pages now describes what it suits.
Websites now get a sitemap.xml and robots.txt automatically.
ClawFlow AI steps now use your own provider key when the agent runs on one.
Integrations
Zenchef: agents and apps can now connect with an API key to read restaurant reservations.
Lightspeed Restaurant: agents and apps can now connect to K-Series with an API key.
WhatsApp: the channel settings now explain why the Save button is disabled.
Models
The automatic model router now picks a model per message on new agents.
For developers
clawnify init --type agent now scaffolds an agent (instructions, skills and flows) as a member of a workspace.
New get_app_source agent tool returns an app's file list, schema and file contents before an update.
The PDF and video render services now accept output: "url" and return a shareable download link.
New screenshot render service turns HTML into an image.
call_app_api now returns a download link for binary responses like PDFs and images, instead of filling the context.
CLI 0.9.1 completes sign-in when the browser opens on the same machine.
Fixes
Fixed VAT not being added at checkout.
Fixed paid organizations showing a trial in the sidebar.
Fixed organization members other than the owner seeing an empty agent inbox.
Fixed newer WhatsApp group IDs being rejected in the group allowlist.
Fixed replies from one chat appearing in another chat that was open.
Fixed recurring skills not loading into the agent.
Fixed creating a sub-agent showing an error when it took a while.
Fixed app drafts showing "no such table" after a database change.
Fixed website images going missing after an assistant edit.
Fixed reconnecting GitHub for a website stopping on GitHub's settings page.
Fixed code blocks in agent replies breaking apart when they mentioned a file or app.
Fixed one heavy task restarting the whole agent.
Fixed long conversations on Clawnify Lite locking up and dropping queued messages.
Week 27
White-label for agencies, invitations people approve on purpose, and a CLI that signs in from any browser.
White-label for agencies
Agencies on the new ServiceOS plans can now serve the Clawnify dashboard on their own custom domain, set up from the dashboard settings.
Teams can now invite people by email or with a copyable link, nobody joins until they accept, and an invite with the owner role hands the organization over on acceptance.
CLI sign-in from any browser
clawnify login now works from SSH and remote sessions, by forwarding the sign-in automatically or showing a code to paste back into the terminal.
Agents and apps can now connect to Fatture in Cloud to work with Italian e-invoicing data.
Forms on websites
Clawnify websites now have a standard Form block, with spam protection built in.
Improvements
Recurring skills now appear as one-tap suggestion chips under the chat box.
People in more than one organization can switch between them from a dropdown in the top-left corner.
The ServiceOS plan can now be bought from the dashboard without a sales call.
The pricing page now lists ServiceOS Pro and a full-width Enterprise plan.
The pricing overview card now shows the lead plan's features and models.
About 240 more API-key integrations can now be connected, and each key is checked with the provider before it is saved.
Long-tail integrations now show their brand logo next to their name.
App builds now stop early when a database change would conflict with the live data, and simple additions like a new column are migrated automatically.
App previews in chat now show the draft while an app is being built and the published version once it is live.
The sidebar now marks the app you have open, and the app name no longer repeats in the top bar.
For developers
New clawnify connections list command shows the exact integration ids, provider key names and custom environment variables your org has, so apps reuse them instead of guessing.
The list_integrations MCP tool now returns the same catalog of integration ids, key names and environment variables.
list_integrations now includes a validated OpenAPI spec link for 12 providers, including GitHub, OpenAI and Stripe.
clawnify login now prints the sign-in URL so you can open it by hand.
clawnify login no longer asks for an organization, and the CLI asks once on the first command that needs one.
@clawnify/connections 0.3.0 adds providers that use HTTP Basic authentication.
The build-a-clawnify-app skill now covers more of the app-building workflow.
The site's llms.txt now tells coding agents how to install the Clawnify CLI and skills, with current pricing.
The company_knowledge_list MCP tool now returns a lean list instead of every document's full text.
Fixes
Fixed being signed out of the dashboard when the login cookie got damaged.
Fixed invitations not recognizing people who already had an account.
Fixed images and other files missing from a website's GitHub repository after sync.
Fixed agents being unable to update or delete app records when the record id made the approval request too long.
Fixed app updates failing with "no such column" when they added a column plus an index on it.
Fixed app drafts using an out-of-date database layout after a schema change.
Week 26
A visual website editor, a real inbox for each agent, and fewer replies that fail halfway.
Visual website editor
You can now edit your website on a device-framed canvas with a Layers panel, a design panel and a built-in chat, and text and theme edits show on the draft site without a rebuild.
A Gmail-style inbox for every agent
Each agent's Email section is now a threaded inbox with Inbox, Sent, All Mail and Rejected folders, where you can open a thread and read full messages and attachments.
ClawFlow HTTP triggers
Workflows can now be started by an incoming web request, and the dashboard builds the example request from the flow's own inputs.
Apps can now schedule a job for later with @clawnify/queue, and the platform handles the timing, retries and signed delivery back to the app.
Fewer replies that fail halfway
When a model returns an empty or failed answer on a demanding turn, the agent now recovers instead of showing an error.
Improvements
Chat now shows "<agent> is thinking" with a wave instead of a spinner.
Slow replies keep the connection alive, so long turns no longer time out in the chat.
Switching back to a chat that is still running now shows that it is running.
Choosing /compact from the slash menu now runs it right away.
You can now drop PowerPoint files into the chat for your agent to read.
Agents now send email with their own email tool.
Agents now tell native channels like WhatsApp, Telegram and Slack apart from integrations, and no longer suggest triggers that do not exist.
The Company Knowledge index agents see is now capped at 30 documents, and you can pin a document or set it to search-only.
Company Knowledge now marks published documents untouched for 90 days as stale.
The main agent can now share one of its recurring skills with a specialized agent.
Copy buttons now work inside apps embedded in the dashboard.
AI-built apps now know how to use your connected integrations.
Clicking a failed agent now shows its error in a popover instead of opening the agent.
Integrations
Meta Ads: insights can now come back as a day-by-day series.
X, LinkedIn, Instagram and TikTok: added as integrations.
Bird: the connect form now asks only for the API key, with the rest filled in automatically.
Models
Auto routing now works when a model from your own API key is selected, and keeps your model when it has nothing to route to.
New models from OpenAI, Anthropic, Google and OpenRouter work as soon as they are released.
With your own single-provider key, image and PDF reading now run on that key.
Demanding turns like building or deciding now get the strongest model in your tier.
For developers
Public routes declared in clawnify.json now get CORS headers and preflight answers automatically.
connect(service).get() and the other HTTP methods in @clawnify/connections now call a provider's own API for any key-based integration with no per-integration code.
Queue deliveries are signed with a public key published as JWKS, and @clawnify/queue verifies them with verifyDelivery().
One-shot CLI commands now exit as soon as they finish.
Website deploys from the CLI work again.
Fixes
Fixed the same message appearing twice after a slow reply.
Fixed subscribing to integration triggers failing.
Fixed webhook events not reaching the agent they were meant for.
Fixed agents being blocked from subscribing to triggers, updating or deleting apps and deleting stored files.
Fixed the main agent's chat history going missing.
Fixed new agents getting stuck while installing.
Fixed agents that were still being set up opening when clicked.
Fixed website sections that fade in on scroll staying hidden.
Fixed the billing portal's back link going to a broken address.
Week 25
New default models on Pro and Max, agents that always know your apps, and chat that keeps the messages you send mid-reply.
A stronger model on Max
The Max tier now runs on a newer flagship model for long, complex work.
Messages queue while your agent replies
Messages you send while the agent is still answering now wait in a queue under the chat and go out one at a time when it finishes, instead of disappearing.
Agents always know your live apps
Agents now see every app your organization has deployed in every conversation, so they can open or update one without asking where it is.
Home projects for specialized agents
Each specialized agent created from the main chat now gets its own Knowledge project instead of sharing the main agent's.
One-step setup for coding tools
Developers can now run clawnify ai-files install to set up Clawnify rules, skills and MCP access for their AI coding assistant in one command.
New specialized agents now start with default persona, rules and tools files, and the main agent briefs them on their role.
The dashboard assistant now hands a task to an agent instead of saying it cannot do it.
The dashboard assistant now runs on your own Anthropic, OpenAI or Google key when your organization has one.
Agents now find the right integration action by searching, instead of loading every action's full details and running out of room.
A stuck integration action now times out after 45 seconds, so the chat no longer looks frozen behind it.
The new agent dialog is wider.
Models
Pro and Lite now answer more reliably when a model is slow to respond.
Requests that include images or files are now sent to a model that can read them.
Automatic model switching now stays off when you turn Auto routing off, even when the router is slow to answer.
For developers
The Clawnify Services PDF render endpoint now accepts width, height, viewport and preferCSSPageSize, so slide decks render at their exact page size.
Fixes
Fixed buying credit packs failing with a "No subscription found" error.
Fixed the usage page showing no usage.
Fixed Slack never connecting after you enabled it as a channel.
Fixed enabling Slack failing on some agents with a configuration error.
Fixed agents sometimes printing their internal reasoning or instructions in a reply.
Fixed adding an API key replacing the primary model you had picked.
Fixed the model picker hiding your chosen model or showing its raw id after a reload.
Fixed chat history and New chat showing another agent's conversations.
Week 24
A front-door assistant on the dashboard, Company Knowledge inside every agent, and GitHub sync for apps and websites.
Dashboard assistant
The dashboard homepage now has a quick-chat assistant that can search Company Knowledge, browse the web and export a CSV, or hand the task to the right agent with a hand-off card.
GitHub sync for apps and websites
Organizations can now keep every app and the website in one GitHub repository, each in its own folder like apps/<slug>/ or website/, and a push redeploys only the folders that changed.
Company Knowledge inside every agent
Agents can now search, read and propose Company Knowledge documents on their own, citing the version they used, instead of only through an outside MCP client.
Import from Claude
Owners can now bring a project from a Claude data export into Company Knowledge from the Knowledge tab, as published documents or drafts.
Pick a model with your own keys
Teams that bring their own OpenAI, Anthropic, Google or OpenRouter key can now pick one of that provider's models for a chat from the model picker and pin it as the agent's default.
WhatsApp conversation reading
Agents can now list your WhatsApp chats and read a conversation.
Improvements
Unused trial credits now carry over when you upgrade to a paid plan.
App databases and file storage are now created in your organization's region, so EU organizations keep new app data in the EU.
Company Knowledge documents now publish with their current edits in one click.
Company Knowledge now indexes PowerPoint files.
You can now delete conversations from the dashboard assistant's recent list.
The dashboard assistant's recent list now includes your agent chats.
Buttons, cards, dropdowns and the organization switcher got a refreshed look.
The new-chat box now sits higher on the page instead of in the middle.
Agents now always see which integrations your organization has connected, so they use them directly instead of searching first.
GitHub connect now lets you pick a personal or organization account, and a Reconnect button names the account when access lapses.
Integrations
Meta Ads: connections now use Meta's own sign-in and refresh the token automatically before it expires.
GitHub: new repositories are created reliably even after your GitHub sign-in token expires.
Models
Lite now replies faster when a model provider is slow.
Thinking levels now work on the Lite, Pro and Max tiers.
When a model fails partway through a reply, the agent now carries on instead of showing a raw provider error.
For developers
clawnify workspace init scaffolds a workspace repository for an organization's apps, and clawnify init places new apps under apps/<slug>/.
clawnify deploy --from <repo> --path <subdir> deploys one folder of a GitHub repository.
MCP connector users who belong to several organizations can now pin a connector link to one of them, and only that link is shown.
The /v1 API now accepts the org clw_ token.
Apps can now render an HTML composition to an MP4 video with the Clawnify Services video render endpoint.
@clawnify/integrations 0.1.1 adds a Meta Ads client with ad account and insights methods.
Website deploys from an archive now keep images and fonts instead of dropping them.
Fixes
Fixed streaming replies pulling you back to the bottom while you read earlier messages.
Fixed integrations whose accounts were removed still showing as connected with no Reconnect button.
Fixed connecting an integration in one organization disconnecting it in another.
Fixed the error banner staying on screen after sending a new message.
Fixed the app split view opening for apps that are not deployed and reopening after you closed it.
Fixed the out-of-credits message showing a raw provider error.
Fixed the dashboard assistant's hand-off card never appearing.
Fixed the message handed over from the dashboard assistant disappearing from the agent chat.
Fixed deployed app previews looping on reload.
Fixed apps embedded in the dashboard not being able to go fullscreen.
Fixed billing checkout errors showing a spinner that never stopped.
Fixed the first message in a new chat failing with a session error.
Fixed agents announcing a next step and then stopping without doing it.
Week 23
Company Knowledge launches, and integrations, secrets and apps got a shared way to reach your credentials.
Company Knowledge
Owners and admins can now keep policies, pricing and brand in one versioned home from the new Knowledge tab, with a review queue, publishing and export, and reach it from Claude and ChatGPT as a custom connector.
Integrations now show an amber Expired or red Error badge when an account lapses or fails, with a Reconnect or Retry button instead of a silent failure at run time.
Secrets in any tool
Agents can now pass a stored secret to any tool by writing {{env:KEY}}, so the value is filled in at run time and never appears in the chat or the saved transcript.
Connections for apps
AI-built apps can now use the organization's connected integrations and secrets through one SDK, @clawnify/connections, without handling credentials themselves.
New agents can now create WhatsApp groups, add participants and share an invite link.
Improvements
The app builder now follows a shared design system, so new apps come out with consistent buttons, numbers, badges and dark mode.
Agents can now look up a saved memory on demand with a dedicated recall tool.
Contact sales buttons now open a scheduling calendar.
The sidebar no longer shows a scrollbar while still scrolling.
Models
Lite now picks the fastest available provider for its model, so first replies arrive in about a second instead of queueing under load.
Managed tiers never send agent traffic to model hosts that may train on it.
For developers
@clawnify/connections and @clawnify/integrations give apps connect(), secret() and describe(), and work with new integrations without an SDK release.
The /v1/connections API lets apps and local tools get tokens, run integration actions and check what is connected, using a Clawnify OAuth bearer or the org clw_ token.
Apps can now run managed integration actions, such as Google Ads queries, scoped to their organization.
The MCP connector now supports OAuth discovery, so Claude and ChatGPT custom connectors sign in with the built-in flow.
Company Knowledge MCP tools let any caller search, read, list and propose documents, while writing and publishing stay with people.
The Deploy button badge for template READMEs now reads "Deploy to Clawnify" with the Clawnify logo.
Fixes
Fixed chats hanging after an interrupted run instead of recovering.
Fixed finished tool steps spinning again when a new run started.
Fixed agents stopping mid-task right after announcing more work.
Fixed the agent's browser view stealing focus from the chat box while a reply streamed in.
Fixed messages the agent started on its own, like scheduled announcements, staying hidden behind the welcome screen.
Fixed template apps built on @clawnify/db working locally but failing with a type error after deploy.
Fixed app deploys failing when schema.sql contained comments with semicolons.
Week 22
Agents get long-term memory, flows get an HTTP trigger, and chat gets better at titles, math and files.
Agent memory
Agents now remember facts across conversations, recall the relevant ones automatically before they reply, and save new ones on their own.
Memory over MCP
Claude Code, Codex, Cursor and other MCP clients can now read, search and write the same memory through the Clawnify MCP gateway.
Any outside tool or script can now run a ClawFlow flow with one POST request and a per-trigger token, which you create, rotate and revoke from the flow's HTTP Triggers panel.
You can now paste text from your computer into the agent's live browser and copy text from it back to your clipboard.
Current docs and web search for agents
Agents can now look up up-to-date documentation for software libraries through Context7, and agents on Clawnify model tiers can search the web.
Improvements
Each chat now names itself with a short title after the agent's first reply, instead of staying on "New chat".
Math formulas written in LaTeX now render in chat.
Chat now accepts HTML and Excel (.xlsx) file uploads.
Agents no longer stop mid-thought on a line that ends in a colon, like "Let me wait for it:".
Every agent's browser now has Chrome's built-in DevTools tools turned on, so agents can inspect the pages they work on.
Integrations
Integrations: agents now see every action a connected app offers, not a partial list.
Integrations: actions now run on the latest version of each integration.
WhatsApp: the channel now comes installed on new agents.
WhatsApp: the same-phone option now describes what it does correctly.
Models
New agents now default to high thinking, so Lite-tier agents show their reasoning too.
For developers
POST /v1/triggers/:triggerId/run runs a flow with a bearer token and returns a run you can poll for status.
HTTP triggers allow 60 runs per minute per token and accept several tokens per trigger for rotation.
The MCP gateway now has memory tools to write, recall, list, get, pin and delete memories.
New /v1/memory API lets signed-in clients write, read, recall, list and pin memories.
Fixes
Fixed the agent's browser tool not working after the latest OpenClaw update.
Fixed new agents missing Clawnify's own tools for apps, storage and integrations.
Fixed some agent tools failing to load on new agents.
Fixed tool steps cutting into the agent's text in chat.
Fixed tool steps not showing in chat while the agent worked.
Fixed markdown symbols showing in the thinking preview.
Fixed chat jumping back to the bottom while you scrolled up to read.
Fixed WhatsApp setup failing on new agents.
Fixed agents being interrupted when a reply held several similar drafts, such as a batch of emails.
Week 21
The Clawnify MCP gateway opens to outside AI tools, website drafts get a live preview, and the CLI and @clawnify/db grow up.
Clawnify MCP gateway
Claude Desktop, Cursor and ChatGPT can now connect to your Clawnify organization through one MCP endpoint, with tools to run code against your apps and search their docs.
Website edits now show in a private draft preview that reloads on its own after each change, and Publish puts the current draft live without rebuilding it.
Public pages and endpoints in apps
Apps can now list routes in clawnify.json that anyone can open without signing in, such as a public GET endpoint while writes stay protected.
Apps can now describe their tables once in schema.ts with @clawnify/db, and the CLI generates and applies the migrations on clawnify dev and clawnify deploy.
Approvals for agent changes
Agents now ask for your approval before deleting an app, changing WhatsApp groups or making other changes through Clawnify's own tools, not just integration actions.
Improvements
Onboarding no longer stops on a waiting screen, and a status pill in the dashboard sidebar shows setup progress instead.
Onboarding has a button in the top left to log out or go back to the dashboard.
Agent cards now say "Online" or "Onboarding" instead of raw status codes.
Edit requests sent to the same app now build one after another instead of colliding.
Apps the agent builds now use React and @clawnify/db by default.
New agents can read PDFs and extract text from documents out of the box.
The website preview shows the live site by default and switches to the draft only while an edit is in progress.
Websites show a loading page while the draft starts up, instead of an error.
Website builds start faster because common dependencies are already installed.
New agents run OpenClaw 2026.5.20.
Models
Requests that hit a provider error are now retried on their own.
Empty replies from a model are now retried automatically.
Picking a Clawnify tier now turns off the Auto Router, and the toggle is hidden while a tier is selected.
The Auto Router is now off by default on new agents.
For developers
GET /v1/mcpb returns a ready-made bundle that connects Claude Desktop to your organization.
Small apps now appear on the MCP gateway as their own tools, named <slug>.<procedure>.
clawnify execute [file] runs TypeScript against your organization through the MCP gateway.
clawnify docs [slug] prints an app's agent-ready README, or the organization's app index.
clawnify mcp runs a local stdio bridge to the MCP gateway for clients without a shell, like Claude Desktop.
MCP gateway calls are limited to 100 tool calls per minute per organization, and every call is written to an audit log.
api.public_routes and api.automation_bypass in clawnify.json declare which app routes are public and allow automations through.
clawnify skills add, clawnify skills list and clawnify skills remove manage the skills in a project.
clawnify init scaffolds schema.ts with @clawnify/db, and --yes accepts every default.
pnpm create clawnify-app (and the npm, yarn and bun equivalents) scaffolds a new app with the Clawnify CLI.
clawnify dev and clawnify deploy now run drizzle-kit for you, so schema.ts is the only schema file to edit.
App deploys now apply drizzle-kit migrations from the drizzle/ folder.
@clawnify/db adds Drizzle support through getDB(), re-exports the common Drizzle operators, and ships drizzle-orm as a regular dependency.
Docs gained an API page covering authentication and rate limits, a concepts section, and an MCP section in the sidebar.
Fixes
Fixed agents failing to read images and PDFs.
Fixed apps deployed from the CLI showing as unpublished drafts.
Fixed the first table in an app's schema being skipped when the schema began with a comment.
Fixed the website draft preview not loading inside the dashboard.
Fixed website builds missing files on larger sites.
Fixed the dashboard hitting a rate limit when loading integrations for several agents.
Week 20
Apps get a draft-then-publish flow and their own database migrations, and chat shows more of what the agent is doing.
Draft and publish for apps
App changes now land in a draft first, marked as a draft in the dashboard, and the live app only changes when you click Publish.
Every deployed app now keeps its own migration history, so the agent can change an app's tables without losing the data already stored.
Agents fix their own failed builds
Agents now wait for an app build to finish and read the error when it fails, so they fix the build instead of reporting success.
App preview beside the chat
Chat now shows build progress while the agent creates or updates an app, and the app opens in a split view next to the conversation when the build finishes.
Context and model notices in chat
Chat now shows a banner while the agent condenses a long conversation and a notice when a backup model takes over mid-reply.
Improvements
The app preview no longer opens on its own when the latest build failed.
The app builder now checks code syntax while it writes, catching errors before the build runs.
Apps and websites shown inside the dashboard can now download files.
Repeated failed attempts in chat now collapse into one note, such as "Agent retried 3× before responding".
Tool steps in chat now appear at the right point in the agent's reply.
Each inbound email now gets its own agent session instead of sharing one ever-growing conversation.
Agents replying to email no longer stop on command approvals that have no chat to appear in.
Keys and environment variables now reach agents at startup without being stored on the agent's disk.
New agents run OpenClaw 2026.5.12.
Prices on the pricing page no longer blur when they change.
Models
Lite, Pro and Max now use fixed model versions, so their behaviour does not change when a provider ships an update.
For developers
New publish_app tool publishes an app's draft to its live address, and update_app now updates the draft only.
New write_migration tool adds a migration file that the next deploy applies inside a transaction.
create_app and update_app now return the build result, including errors.
clawnify deploy now runs without prompts, so agents and CI can use it.
clawnify deploy now writes the deployed schema.sql back to your project, and clawnify pull schema --from <id> fetches it on demand.
New @clawnify/db package gives apps one SQLite client that works in both the draft and the live app.
Keys an app declares under env in clawnify.json are now filled from the organization's custom environment variables as well as provider keys.
Apps installed from a template now record the template's repository and commit.
Clawnify CLI 0.1.17.
Fixes
Fixed the "+ New agent" button not opening the new-agent dialog.
Fixed the Stop button not always stopping the agent's reply.
Fixed chats looking busy or finished at the wrong time.
Fixed chat names clashing with another chat's name or coming out blank.
Fixed the agent's browser refusing to open websites by address.
Fixed new agents getting stuck while installing.
Fixed provider keys saved before the switch to organization settings not reaching agents.
Fixed draft app previews returning a 403 error.
Fixed some app builds failing to start.
Week 19
Clawnify Websites launches with custom domains, alongside a PDF service for apps and new ClawFlow steps.
Clawnify Websites
Every organization can now have a public Astro website that its agent builds and edits from plain instructions, hosted at a myclawnify.com address.
Custom domains for websites
Owners can now connect their own domain to the website, and connecting the root domain sets up the www version too.
Preview first, then publish to live
Once a website has a custom domain, agent edits land in a live preview that updates as the agent works, and visitors see them only after someone clicks Publish to live.
PDF rendering for apps
Apps can now turn HTML into a PDF through a Clawnify-managed service, using the @clawnify/pdf package and the organization's Clawnify token.
App and integration steps in ClawFlow
ClawFlow workflows can now call your Clawnify apps and run integration actions as built-in steps, instead of hand-written code with its own authentication.
Website builds show a live status line under the badge, such as "Editing Hero.astro", while the agent works.
Website edits now build on the site's existing source files, so earlier changes carry over to the next edit.
Websites are styled with Tailwind v4 and a DESIGN.md design brief, and pages are composed from flexible layout blocks instead of fixed section types.
Website previews show in the dashboard next to the build status.
A custom domain goes live as soon as one of its hostnames validates, and setup lists every certificate TXT record to add.
A refresh button next to a custom domain rechecks its status on demand.
Environment variables are now set for the whole organization and apply to every agent.
The Website link moved into the Account section of the sidebar.
Models
Requests now retry on their own when a model provider has a temporary error.
Model errors in chat now include the provider's explanation instead of only a status code.
For developers
clawnify link connects a local folder to an existing app, and clawnify deploy then redeploys that app.
clawnify deploy asks whether to update the existing app or create a new one when the app name is already taken in your organization.
clawnify init scaffolds apps with React, Tailwind v4, OpenAPIHono and zod-openapi.
New @clawnify/pdf package renders HTML to PDF from an app, with self-hosted fallbacks.
Each organization now has one Clawnify token for every Clawnify service, available to deployed apps as CLAWNIFY_TOKEN.
update_website now takes a single plain-language instruction, on the MCP tool and on the CLI.
New publish_website tool and clawnify publish-website command promote a website preview to live.
New upload_asset tool returns a permanent public URL for an uploaded image or file.
Fixes
Fixed the first message flickering when sent from a new chat.
Fixed saving a custom environment variable failing with "Failed to save metadata".
Fixed provider keys saved in Settings not reaching newly created agents and newly built apps.
Fixed website deploys failing.
Fixed connecting a custom domain failing.
Fixed reconnecting a domain that had already been added returning an error.
Fixed custom domains showing as failed while DNS was still propagating.
Fixed custom domain status not updating after the domain was verified.
Fixed websites on custom domains not loading.
Fixed the app list returned by clawnify ls.
Week 18
Easier setup for extra agents and triggers, saved skills one click away, and a link for every chat.
New agent form
Owners can now add an agent from a + New agent form by giving it a name, a description and a level, and it starts working alongside the main agent.
Saved skills under the chat box
Tasks your agent has saved as reusable skills now show as cards below the composer on every new chat, so you can run one with a click and the chat shows a Running skill chip.
Instructions for each trigger
Owners can now tell the agent what to do for each trigger in a What should the agent do? field, and choose whether the trigger connects to the agent right away.
Describe the event, get the trigger
The trigger picker now starts with a text box, so you describe what to react to in your own words and it finds the matching trigger and fills in its settings.
A link for every chat
Every chat, including a new one, now has its own address, so you can bookmark it, refresh it or open it in a new tab without losing your place.
Improvements
Your first message in a new chat now goes straight to the agent instead of being preceded by a greeting.
The + New chat button and chat history items are now real links, so they open in a new tab with a modifier click.
The trigger dialog is wider and lists available triggers in three columns.
Inbound emails now arrive with your custom email instructions in the same message, and the agent replies by email rather than in chat.
Apps can now use integrations that connect with an API key or a username and password, not only ones that sign in with an account.
Agents now run workflows on ClawFlow 1.0.
The Clawnify website now offers a free AI Apocalypse Score that rates how exposed a business's website is to AI, across five categories.
Integrations
WhatsApp: agents can now create groups, add people to them and share a group invite link.
WhatsApp: new links now only accept direct messages from allowed numbers by default.
For developers
clawnify deploy now redeploys the linked app, and when a name is already taken it asks whether to update that app or create a new one.
clawnify init now scaffolds a React and Tailwind v4 app.
Scaffolded projects now pin clawnify as a dev dependency and require Node.js 22 or newer.
clawnify dev and clawnify deploy now stop with a clear message on Node.js versions older than 22.
Apps that declare API key or username and password integrations receive them as secrets at build time.
The API now publishes OAuth protected resource metadata at /.well-known/oauth-protected-resource, so agents can discover how to sign in.
The dashboard now exposes listing, reading and creating apps and listing agents as WebMCP tools for browser agents using your session.
Fixes
Fixed approving or declining an action in chat stopping the agent mid-task.
Fixed linking WhatsApp failing with a "web login provider is not available" error.
Fixed enabling Slack and Telegram failing on older agents for the same reason.
Fixed the self-chat setting being ignored when first linking WhatsApp.
Fixed clawnify dev failing to find the app's backend code in new projects.
Week 17
Agents can now react to events in connected tools, take actions in them, and ask for approval in chat first.
Triggers from connected integrations
Owners can now turn on triggers from a connected integration's card on the Integrations page, like a new Slack message or a Google Sheets change, and the event goes straight to the agent with no workflow to set up.
Set up a trigger in your own words
Instead of a settings form, the trigger dialog asks one plain question, like which sheet and range to watch, and turns your answer into the trigger's settings, with the full form still under Configure manually.
Agents act in your connected tools
Agents can now run actions in connected integrations, like reading a spreadsheet or posting a message, and the chat shows the integration's icon on each action call.
Approvals in chat
When an action needs sign-off, the agent now asks in the chat you are using and waits, and you can allow it once, allow it always or deny it.
More than one agent in a workspace
Your agent can now create additional agents, each with its own chat, and they show as cards on the main agent's home screen.
Improvements
Inbound emails and trigger events now reach the agent's main conversation, with the full content saved to a file the agent opens when it needs it.
Workflow triggers can now send events to the agent directly, to a ClawFlow workflow, or to both.
The chat composer now sits centered on a new chat and moves to the bottom once you send the first message.
An approval request now waits up to nine minutes for an answer before the action is dropped.
The Clawnify website now offers a free email signature generator.
For developers
Agents on Clawnify can run any integration action with clawnify action <SLUG> and look up valid action names with clawnify list-actions --toolkit <slug>.
When an integration action fails, the CLI now shows the integration's own error so the agent can correct the request.
New MCP tools create_agent and list_agents let an agent create and list the other agents on its machine.
The API now has /v1/policy/check to decide whether an action is allowed, denied or needs approval.
Fixes
Fixed creating a trigger from the Integrations page failing.
Fixed integration actions failing to run when called by name.
Fixed filtering available actions by integration.
Fixed trigger events from connected integrations being rejected as unsigned.
Week 16
Two-way GitHub sync for apps, integrations shared across the organization, a billing page and model tiers matched to plans.
Two-way GitHub sync for apps
Owners can now link any app to a new or existing GitHub repository from the GitHub icon in the app header, so a push redeploys the app and every Clawnify build commits its source back to the repository.
Integrations shared across the organization
Owners can now connect an integration once from Settings → Integrations and every agent in the organization can use it, while a single agent can still be given its own account that takes precedence.
Billing in settings
A new Settings → Billing page shows your current plan and credit balance and lets you upgrade, downgrade or choose a plan, and it sits next to API Keys in the account menu.
Connected integrations inside apps
Apps can now use the organization's connected integrations, like Google Workspace, by declaring them in clawnify.json, and they get fresh tokens without storing any secrets.
Owners can now add WhatsApp groups one at a time and choose per group whether the agent replies to every message or only when mentioned, instead of an allowlist that silently turned groups off.
clawnify dev
Developers can now run an app locally with one clawnify dev command, which sets up the local database and dev environment from clawnify.json.
Custom environment variables now have their own Settings → Environment Variables page, separate from API Keys.
The New chat button now starts the agent's greeting, so it reads its persona and memory before you type.
Chats no longer jump to the bottom on every new agent message.
Only organization owners can now start a checkout for a plan.
Canceling checkout now returns you to onboarding with a notice instead of a blank screen.
The extra Start free trial button at the bottom of the onboarding plan step is gone, since each plan card has its own.
The app builder now picks a template from clearer descriptions, asks when a request is ambiguous, and builds from scratch instead of forcing a weak match.
New custom apps now start from the same scaffold as clawnify init, so the builder customizes a known base instead of generating everything.
Apps imported from a repository keep their own build script.
GitHub sync leaves build output out of the repository.
Models
Automatic model routing is now off by default.
For developers
clawnify init now takes --name, --template and --storage flags for non-interactive use.
clawnify init now writes a fuller .gitignore.
Apps declare the environment keys and storage they need in clawnify.json.
Apps declare the integrations they use under app.credentials in clawnify.json, documented in the app template reference.
The docs now cover the identity headers apps receive and the Clawnify model tiers and routing.
Fixes
Fixed accounts without a subscription being sent away from the dashboard.
Fixed checkout errors failing silently, so they now show a message and can be retried.
Fixed an agent being created before checkout had completed.
Fixed the credit amount resetting to the default when starting a trial.
Fixed agents' instructions growing longer with every turn of a chat.
Fixed the GitHub connect popover redirecting again when the GitHub app was already installed.
Fixed the WhatsApp channel failing to start when saved with an empty allowlist.
Week 15
A rebuilt app builder, credit-based pricing, the first Clawnify CLI and a public docs site.
A more capable app builder
The app builder can now take on larger, multi-file apps instead of generating them in a single pass.
Plans with a credit balance
Every plan now comes with a monthly credit balance you pick from a dropdown, with the price updating as you choose.
Developers can now scaffold, run and deploy apps from the terminal with npx clawnify, and use the same commands from Claude Code through its built-in MCP mode.
Apps opened from the dashboard now receive the signed-in user and their organization, so they can personalize content and control access without a separate login.
Onboarding is now four steps (Setup, Role, Plan, Review) instead of six, with channels and API keys set up after sign-up.
The pricing page was rebuilt to match the onboarding plan cards, with a monthly and yearly toggle.
Yearly plans now show "billed yearly" instead of a calculated annual total.
Pricing page sign-up buttons now open the sign-up page in a new tab.
Credit packs are now offered in three sizes.
Onboarding no longer includes a meeting step.
When your agent needs approval to run a command, it now pauses and waits instead of queuing follow-up messages.
People who open Clawnify inside the LinkedIn, Facebook or Instagram browser can now sign up with email.
One connected event source can now start workflows on several agents in the same organization.
App builds now upload and unpack files in parallel, so they finish sooner.
Models
Clawnify Pro is now the name of the standard model, alongside Clawnify Lite and Clawnify Max.
Clawnify models now catch common slips, like stopping early or running a destructive command without confirming, and correct themselves mid-reply.
Each Clawnify tier now adapts to the task, giving harder work more capability.
For developers
clawnify init scaffolds a new app from a blank or CRUD template, ready to run locally with Vite and a local database.
clawnify login signs in through a browser consent page and asks which organization to use.
clawnify deploy, ls, logs, open, rm and whoami manage apps from the terminal.
clawnify --mcp exposes deploy, list, get and delete as MCP tools for Claude Code.
The API now ships an OpenAPI spec with an interactive reference at docs.clawnify.com.
The docs site now has a site-wide /llms.txt index and a View llms.txt option on every page.
The docs now include an app template reference covering the file structure, the database pattern and a migration guide.
Apps receive X-Clawnify-User-Id, X-Clawnify-Org-Id, X-Clawnify-User-Email and X-Clawnify-Caller headers, and any copies sent by the browser are stripped.
Fixes
Fixed the send button being blocked while a chat still looked busy after the agent had finished.
Fixed agents repeating the same paragraph within a single reply.
Fixed buying credits failing for organizations created before credits existed.
Fixed the app link missing after deploying an app from the CLI.
Fixed the pricing page scrolling sideways.
Week 14
Integrations can start workflows, chats get a history sidebar, and files, apps and approvals show up inside the conversation.
Integration triggers for workflows
Connected services such as Slack and GitHub can now start ClawFlow workflows through triggers you turn on for the integration, with no webhook URL to paste.
Outside services such as WhatsApp Business through Bird can now start a ClawFlow workflow with a webhook, and the event arrives as the workflow's input.
You can now sign in with a magic link sent to your inbox, as well as with Google.
Improvements
Owners and admins can now rename the organization.
The chat shows when the agent is reasoning, and you can expand it to read its thinking.
A spinner replaces the blinking cursor while the agent is replying.
Tool calls in chat now show the logo of the service they use.
The options under the chat box now have tooltips.
Each integration now has a short description.
The integrations grid now adapts to smaller screens.
Your own OAuth credentials for an integration now go in a separate dialog.
Connecting an account you have already connected no longer creates a second connection.
Agents can now call any connected service's API, not only the services that have a dedicated tool.
Agents read a skill's instructions before using a tool they have not used before.
ClawFlow now comes installed on every new agent.
The pricing page now shows live prices.
If a payment is past due, you see a payment prompt in place of the dashboard.
Onboarding no longer lets you hire another agent once your plan's agent limit is reached.
The feature sidebar is hidden while you set up a new agent.
The organization selector shows your organization's initials instead of the Clawnify logo.
After signing in you now land on the dashboard instead of onboarding.
Models
Auto Router now detects which provider you use and picks between its fast and strong models, such as Haiku and Sonnet for Anthropic.
Every model is now available in /model, with no allowlist.
Fixes
Fixed tool calls appearing out of order while a reply streamed.
Fixed tool calls and the running state not showing after you reloaded the page.
Fixed the running spinner vanishing while the agent was thinking.
Fixed words splitting at random letters in chat bubbles.
Fixed accounts with a past-due payment being sent back to onboarding.
Fixed checkout creating duplicate subscriptions.
Fixed the integrations page showing no integrations after a brief connection problem.
Fixed Google Workspace commands failing with a 403 error.
Fixed the pricing page mixing currencies.
Fixed the dashboard misbehaving when your session expired instead of asking you to sign in again.
Week 11
Every agent gets an email inbox, teams move into shared organizations, and the App Builder starts fixing its own builds.
An email inbox for every agent
Every agent now gets its own email address and an inbox in the dashboard, and it replies in the same thread while keeping track of who was CC'd.
Organizations
Teams can now share agents and apps inside an organization and switch between organizations from the sidebar.
An App Builder that fixes its own builds
The App Builder now runs an agent that builds, fixes and redeploys an app until it works, and new custom apps use Vite, TypeScript and Preact instead of plain JavaScript.
Google, Microsoft 365 and Dynamics 365 integrations
Agents can now work in your Google, Microsoft 365 and Dynamics 365 Business Central accounts without ever seeing your passwords.
Model picker, thinking level and Auto Router
You can now choose a model with the /model command, set how hard the agent thinks from the chat toolbar, and turn on Auto Router to pick a model for each message.
File uploads in chat
You can now send PDFs and other documents to your agent straight from the chat, in addition to images.
Improvements
The team overview page shows every agent along with the integrations each one has connected.
Agents now have their own section in the sidebar instead of a dropdown menu.
Each agent has its own workspace with its browser view, calendar, keys and integrations.
Outbound agent email is checked against an allowlist and goes to the owner when no recipient is allowed.
You can give an agent custom instructions for how it handles email.
In WhatsApp open groups, agents now reply only when someone mentions them.
The chat shows a reconnecting state while the agent restarts.
Model errors in chat now show the provider's actual message and the model name.
The / commands hint now opens the slash command menu when you click it.
A new app is marked live after its first successful deploy, so you can preview it right away.
Apps stay live while an update is being built.
The template matcher asks a clarifying question when an app request could fit more than one template.
The sidebar shows an app's build status while it builds.
Agents now recover from browser errors on their own instead of giving up after a timeout.
The agent sidebar is collapsed by default on mobile.
The trial credits card is hidden once you add your own API key.
Integrations
Google: connecting an account now uses a pre-verified sign-in that covers Drive, Gmail, Calendar, Sheets, Docs, Tasks and Contacts.
Dynamics 365: agents can now work in Business Central, and the connect card shows hints for using your own OAuth client.
GitHub: agents can now use the GitHub command line tool with your connected account.
Integrations: the list of available integrations now loads live, with each service's logo.
Models
Gemini 3 Flash Preview is now the default model for new agents.
The default thinking level is now medium.
Auto Router now picks a model for each task on its own.
For developers
App templates now publish an OpenAPI spec, and agents read it with the get_app_openapi MCP tool, which replaces get_app_endpoints.
Fixes
Fixed line breaks disappearing from chat messages.
Fixed Safari zooming in on the chat box on mobile.
Fixed the message box not staying at the bottom of the screen on mobile.
Fixed the chat losing track of when the agent had finished replying.
Fixed the Auto Router toggle disconnecting the chat from the agent.
Fixed custom apps returning a 404 page after they were deployed.
Fixed apps intermittently failing to load with a 530 error.
Fixed app builds failing silently or skipping the database setup.
Fixed agents receiving or sending the same email twice.
Fixed the agent's browser showing websites in German.
Fixed connecting an integration failing partway through sign-in.
Week 10
A $5 credit trial, keys in one place, Microsoft sign-in and a clearer view of what the agent is doing.
$5 of free credit to start
New accounts now start with $5 of usage credit instead of a 3-hour trial, with the balance shown in the dashboard.
Granola actions and triggers 
Microsoft Teams channel 
Clawnify plugin for Claude 
Clawnify on WhatsApp 
Google Business Profile integration 
GitHub sync for the whole workspace 
Voice notes and spoken replies