LoginSign Up
← Back to all sparks
A

AutoGPT

AI-ASSISTANTS
Velocity5.0

Open-source autonomous AI agent platform that chains GPT outputs to accomplish goals.

agpt.co ↗

AutoGPT bets on an AI staff model — Experts marketplace deepens every week

agentic-platformexpert-marketplacevoice-aiai-workforcedevtools
◆Current state
AutoGPT has pivoted from a freeform agent framework to a platform where you hire AI experts — preconfigured agentic personas with isolated memory, scoped integrations, and dedicated thread history. The 0.7.x series ships weekly, adding scheduling, voice briefings, per-expert spend tracking, and now integration-level isolation per expert. Auth was replaced (Supabase to Better Auth) and a single-source LLM catalog now supports models including Claude Sonnet 5.
◆Where it's heading
Each release deepens the Expert abstraction: tighter control over what each expert can access, more visibility into their activity, and more structure in how they communicate. The activity event log and per-expert integration scoping in 0.7.4 hint at the next logical step — org-level dashboards for managing an AI staff roster, not just configuring individual agents.
◆Prediction
Billing and credit allocation per expert are the near-term missing pieces. Cross-expert task delegation would complete the AI team model — expect a feature in that direction within two or three releases.

◆Recent moves

  1. 11d ago

    AutoGPT 0.7.4: Per-expert integration scoping and homepage activity log

    0.7.4 locks down what each expert can see and do — integrations are now scoped per expert rather than shared globally. The homepage surfaces an activity event log so you can see what your AI experts have actually been doing. Both changes make the Expert model operationally useful rather than just conceptually interesting.

    View source ↗
  2. 19d ago

    AutoGPT 0.7.3: Structured notification design and expert team tooling

    0.7.3 replaces ad-hoc email notifications with a structured Briefing/Alert/Verdict/Ops design — four notification types with clear intent. Live compaction progress in Copilot chat and selectable ask_question options reduce expert setup friction.

    View source ↗
  3. 26d ago

    AutoGPT 0.7.2: AI voice briefings, writing style capture, per-expert spend

    AI voice narration in morning briefings and per-expert spend tracking on the home screen make the Expert experience more practical as a daily workflow tool. Writing-style capture during hire flow — where the platform learns how you communicate — is an onboarding detail that compounds over time.

    View source ↗
  4. 1mo ago

    AutoGPT 0.7.1: Claude Sonnet 5 support and expert scheduling

    Adding Claude Sonnet 5 to the unified model catalog and shipping expert scheduling (attribution, triggers, thread posts with credit guardrails) extends the platform in two directions simultaneously: better models and more autonomous expert operation on a schedule.

    View source ↗
  5. 1mo ago

    Preview seed fixture (rolling)

    An internal preview-environment seed fixture — a synthetic database snapshot restored before each PR's migrations run against populated tables. Invisible to end users; a developer experience improvement for the team shipping the platform.

    View source ↗
  6. 1mo ago

    AutoGPT 0.7.0: Experts marketplace launches with auth rebuild

    ⚡ SPARK

    0.7.0 is the release that changes what AutoGPT is: the freeform agent builder becomes a marketplace of hireable AI experts — each with isolated memory, dedicated threads, and a defined capability surface. Supabase Auth was replaced with Better Auth in the same release. Everything in 0.7.x builds on this foundation.

    View source ↗