Release notes

What's new in Agent Anything

Here are our latest updates and what's coming next.

v1.18.6
October 2026

Steadier conversations, files, and security

  • Clearer conversation status as an agent works: Ready, Gathering, Saving, Attaching, Checking, and Finishing.
  • Conversations can carry far more files forward, so long, document-heavy work no longer runs into a combined size limit.
  • Files save privately while an agent works, with a visible waiting list and clear saved and attached states.
  • Larger downloads from Outputs stream reliably and cancel cleanly if the connection drops.
  • Fixed: a conversation could lock up when Stop arrived just after an answer finished, and quick follow-up requests now recover properly.
  • Fixed: interrupted saves and attachments recover their real outcome without repeating the action.
  • Fixed: agents keep working when a model provider adds new details to its responses, and new conversations prepare reliably with the latest skill versions.
  • Updated to the latest Node.js long-term support release, including the October security fixes for compression, secure connections, and trusted certificates.
  • Patched the document renderer and the components that display formatted text, math, and diagrams, so shared content is cleaned safely before it appears.
  • Tighter limits and time-outs on incoming requests from connected sign-ins and Slack, with stronger checks that each message is genuine.
v1.18.0
October 2026

When we say anything, we mean Anything

  • Agents can now connect to any external system with their own private API credentials, stored securely and activated deliberately. No MCP support required on the other side.
  • Clearer status for background work, separating interactive chats, task services, and scheduled runs.
  • Finished scheduled runs that still need attention stay visible and recoverable.
  • Conversations and tool work settle more reliably, with steadier checkpoint progress.
  • Credential management is quieter, with clearer choices for search history and resume.
v1.17.0
September 2026

Approvals, schedules, and images

  • New and enhanced chat and inspector layout, with clearer avatars and schedule overlaps shown in advance.
  • Image generation and editing are available to agents, configured per agent with a usage allowance.
  • Scheduled runs check for results on a shorter interval, so follow-ups arrive sooner.
  • Approved connector actions pause for a fresh approval when permissions or sign-in get in the way.
  • Uploads are validated before processing, so unusable files are rejected immediately.
v1.16.0
September 2026

A stronger conversation and recovery experience

  • Conversation controls are clearer across History, Continue, Branch, Parent, Active chat, Refresh, Full chat, and Look up.
  • Prior generated work, files, and conversation outputs are easier to find and reopen.
  • Agent responses render as open prose, with a consolidated Chat menu, quieter entry motion, and a more disciplined streaming state.
  • Connector status and actions now live inside the composer, with a compact summary view and an expanded chip view that stays stable as live status arrives.
  • Expanded chat opens with the Agent Profile pane visible, so people can see an agent's capabilities while chatting.
  • Bridge memory answers let connected assistants answer focused questions from authorized memory with cited sources.
  • Scheduled-run audit visibility shows what ran and what happened.
  • The composer accepts multiple file uploads, and duplicate same-name attachments no longer create stale blockers.
  • Approval and recovery flows around connector actions, paused work, and interrupted turns are more resilient, and approvals settle cleanly after terminal provider evidence.
  • Approval tool inputs are redacted from chat transcripts, and the release image ships patched packages with a passing vulnerability gate.
v1.15.0
September 2026

A clearer view into what happened

  • Administrators can expand audit records to see who acted, on what surface, with what outcome, and under which policy.
  • Activity references stay privacy-bounded and indirect, so people only see what they are allowed to see.
  • Nested MCP approvals now hand ownership cleanly to the next approval in the same durable turn.
  • Bridge OAuth confidential-client exemptions are operator-allowlisted, so strict challenge/verifier symmetry still applies everywhere else.
v1.14.0
August 2026

Every team agent, from your phone

  • Open Agents from the phone chat header to find every team agent you're authorized to use, grouped and searchable with status and attention indicators.
  • Switch conversations without stopping background work; unsent text stays with the agent you wrote it for.
  • Reviewers can resolve bounded governance items directly from mobile, without gaining wider access.
  • Read-only access stays honestly view-only, and access changes take effect immediately everywhere.
v1.13.2
August 2026

Steadier managed connections

  • Connections to your own tools stay put across restarts when nothing about them has changed.
  • Unchanged behavior for existing connectors, with fail-closed defaults made explicit.
v1.13.1
August 2026

Replies that stay put

  • A streamed reply now remains visible once the turn completes, with no refresh needed.
  • More resilient recovery when a conversation's live connection is interrupted.
v1.13.0
August 2026

Sharing, guides, and diagrams

  • Team sharing publishes exactly the note you staged, with more reliable turn recovery.
  • Assistants answer from the shipped Help guides instead of duplicated product knowledge.
  • Thinktank notes collapse when long and render diagrams inline.
  • Share up to five files from a conversation at once.
v1.12.0
August 2026

Answers as they're written

  • Chat now streams responses token by token instead of waiting for the finished message.
  • A live connection carries updates to the browser, with polling kept as a recovery path.
  • Every send is claimed durably, so an interrupted response can never be sent twice.
v1.11.1
August 2026

Stop means stopped

  • Stop is authoritative across chat, fleet, and the conductor: Working and Stopping stay visible until the turn truly settles, through navigation, reload, and switching to your phone.
  • Work that can't be stopped reports its status honestly instead of offering a Stop that does nothing.
v1.11.0
August 2026

Files with your scheduled results

  • Scheduled private runs can deliver up to five generated files with their Slack results, with a text fallback if delivery isn't possible.
  • Connector problems now say clearly whether you can retry yourself or need an administrator.
v1.10.x
August 2026

A calmer chat surface

  • Share and second opinion sit together in one compact action row under the latest response.
  • A wider conversation option, a one-click jump back to the latest message, and clearer language for continuing past conversations.
  • Suggested shares recognize real updates and stay quiet on questions, retractions, and hypotheticals.
v1.9.1
August 2026

Improvements to long-running agentic workflows and integrations

  • Agents handle longer workflows more smoothly from start to finish.
  • Integrations stay reliable and keep you informed about their status.
  • Workflow state is preserved more consistently across sessions.
v1.9.0
August 2026

Inline audio in the artifact panel

  • Audio files now play directly in the artifact panel on desktop for speech-to-text and other audio workflows.
v1.8.5
August 2026

Steadier connections to your own tools

  • More flexible connector creation and debugging.
  • Safer connection validation with a narrow reviewed exception.
  • Dismissible reconnect reminders for the session.
  • Clearer rejection of malformed file uploads.
  • Tool approvals only unlock on confirmed results.

Newly verified connectors

Tested in customer deployments. The list keeps growing.

  • Mercury Bank
  • Firecrawl
  • Cartesia
  • Stripe
  • Dynamic
  • Attio
  • Gmail
  • Google Drive
  • SharePoint
v1.8.0
August 2026

A consistent look across every surface

  • Set your preferred appearance to System, Light, or Dark, and it follows you across sessions and devices.
  • The theme carries through the app, sign-in flows, portals, mobile views, and diagrams without changing the original look of authored documents, PDFs, images, or prints.
  • Dark mode provides reversed brand assets, stronger selected states, and a consistent treatment across desktop and mobile.
v1.7.0
August 2026

Connect your own tools, under policy

  • Add your own approved tools to agents without engineering help, using standard sign-in or key-based authentication.
  • Credentials are held in a protected vault and scoped to the right level of access.
  • Every connection carries a clear policy, so agents only act when the rules allow it.
  • Setup and lifecycle issues fail closed and surface as clear, bounded guidance.
  • Help content and operator guidance updated to cover connectors end to end.
Earlier releases
  • v1.6.2August 2026

    Thinktank sharing returns semantic receipts and preserves Markdown, the chat composer accepts pasted images, memory integrations use Memory Box branding, and iPhone chat gets compact chrome and a sticky identity row.

  • v1.6.1July 2026

    Generated artifact versions are immutable and delivered once, uploaded attachments open natively from the Library, HTML artifacts include static Preview and Source views with scripts disabled, and Fleet card layout is more resilient.

  • v1.6.0July 2026

    Diagrams in conversations render as clean visuals and print correctly, PDFs open directly in the app, any artifact exports to a matching PDF, and artifact viewing, downloads, and sharing are more reliable.

  • v1.5.0July 2026

    A clearer, more flexible workspace: a refreshed chat experience, live model conferences in a dedicated theater, a resizable chat and artifact split, and one icon for the full witness receipt.

  • v1.4.2July 2026

    Fixed an issue that could interrupt a conversation before it finished settling.

  • v1.4.1July 2026

    A rewritten in-app Help library, a new guide to model consultations, and more dependable scheduled delivery.

  • v1.4.0July 2026

    Governed model consultation and a real working surface: agents consult approved models under policy, bounded conferences with synthesis, and private-by-construction sessions and files.

  • v1.3.1July 2026

    Scheduled times now state their own timezone, so a schedule always means what you intended.

  • v1.3.0July 2026

    Scheduled runs deliver their results where they belong: to shared memory and to the team feed.

  • v1.2.3July 2026

    The running release version is now visible in-product on the Help page.

  • v1.2.2July 2026

    Work with several agents at once: switch agents mid-turn, and live activity survives navigation.

  • v1.2.1July 2026

    An app-like mobile experience, automatic catch-up after backgrounding, and clearer guidance on how information flows between agents and the shared commons.

  • v1.2.0July 2026

    Slack identity linking for organizations with more than one email domain.

Roadmap

What's next

Where we're heading. Timing moves as we learn from the organizations running Agent Anything today.

In progress

Voice mode for your personal assistant

Talk to your private assistant naturally, with hands-free input and spoken responses that stay within your isolated deployment.

In progress

Richer Thinktank briefings

More useful scheduled briefings drawn from your shared memory, tuned to the questions each team actually asks.

Exploring

Broader governed agent actions

More consequential actions an agent can take under policy, each producing a verifiable receipt.

Want a deployment of your own?

We'll map your current AI surfaces, identify a first use case, and put Agent Anything to work for you and your team.

Get in touch