Skip to content

Changelog

What we shipped.

The big releases, newest first. The docs changelog has the detail, fixes included.

  1. v3.1Latest

    The agent catalog, rebuilt

    11 changes

    • Agent template gallery grown to 77 ready-to-install agents, 20 of them new. Every template now states its guardrails and when the job is done
    • 74 built-in Skills, each a step-by-step procedure that lists the mistakes to avoid
    • Draft with AI now builds the agent it describes: its Skills attached, the Integrations it needs listed for setup, and instructions you can edit before you create it
    • Website chatPut a customer-facing agent on your site with one script tag. It is read-only by default, a daily credit cap limits what it spends, and its Numbers page shows the last 30 days of conversations
    • On GitHub, Jira, Confluence, Notion and Linear, your agent gets exactly the abilities you tick
    • Every setting for an agent on one page
    • Watch each helper work, and see what it costs
    • See what your agent remembers, correct it, or switch memory off
    • Undo a file your agent changed, and give each conversation its own branch
    • Organization-wide approval rules and a minimum approval level
    • A run that reaches its step limit says what is done and what remains, and failed runs explain what went wrong in plain words
  2. v3.0

    The loop era

    10 changes

    • GoalsGive an agent a goal and it keeps taking passes until a separate check agrees it's met — or your budget says stop
    • LoopsAgents book their own next pass and wake on events instead of polling; every pass carries hard time and credit budgets
    • HelpersAn agent can split a wide job across helpers that work in parallel, then write one answer from what they return
    • Self-evolving memoryAgents write and consolidate their own notes and reflect after every run, so the next pass starts smarter — and every entry is yours to read and edit
    • Voice rebuilt on LiveKitOn a call in the app, hand over a job by voice and hang up, ask for progress from the live run, watch artifacts render mid-call, and hear helper milestones as they land
    • Bring your own machineEnroll a Linux box you own with one command, over an outbound-only connection
    • StudioAgents and Skills drafted by AI from a description, a chat thread, a screen recording, or a YouTube link — always to the editor for your review
    • An autonomy dial on every agent — Autopilot, Guarded, Supervised, or Co-pilot — and a scheduled pass obeys the same dial you set
    • Fork any conversation to explore a branch without losing the original
    • Agent template gallery grown to 58 ready-to-install agents
  3. v2.0

    The premium relaunch

    8 changes

    • Sandbox co-workingThe agent's terminal, browser, and files are open to you mid-run — take control of the browser, hand it back
    • Long autonomous runs with state checkpointed after every step; interrupted runs resume with full state
    • VoiceReal phone calls via Twilio in designed or cloned voices, narrated decks, and podcasts
    • Model choice per agent
    • Featured MCP catalog (28 servers, vendor-hosted and self-hosted) plus custom MCP server support
    • Google Workspace CLI in every sandboxGmail, Drive, Docs, Sheets, Calendar
    • Agent template gallery25 ready-to-install agents
    • PresentationsPrompt to deck, theme designer, export to PDF/PPTX/MP4, publish to a link
  4. v1.0.0

    Initial Release

    6 changes

    • Adaptive multi-agent orchestration with specialized agents for planning, execution, and verification
    • Computer Use for vision-based UI automation
    • Knowledge base with RAG-powered document retrieval
    • Flows for event-driven and scheduled automation
    • WhatsApp, Telegram, and SMS integrations via Twilio
    • GitHub, Jira, Confluence, and Google Drive integrations

  5. Fixes and smaller changes are in the docs changelog.

    Open the docs changelog