LoginSign Up
← Back to all sparks
A

AutoGPT

AI-ASSISTANTS
Velocity7.5

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

agpt.co ↗

AutoGPT is building a workforce: experts now get schedules, credits, and their own briefings.

agent-platformexpert-schedulingproactive-agentsmarketplacellm-integrations
◆Current state
The last three releases all advance one idea. v0.7.0 split the Copilot into experts with scoped sessions, identity context and a marketplace, on a rebuilt Better Auth foundation. v0.7.1 gives those experts schedules — attribution, triggers, thread posts and a credit guardrail — plus editable Soul documents, collapsible expert chat groups in the sidebar, and a briefing-first home built around a morning briefing and unified needs-attention view. Tavily search/extract/crawl/map blocks and Claude Sonnet 5 support land in the same release. Underneath, v0.6.69 had already taught the copilot bot to post into Slack and Telegram unprompted.
◆Where it's heading
The platform is converging on persistent, scheduled, individually-billed agents that report back rather than wait to be asked. Scheduling with a credit guardrail is the piece that makes that economically safe; Soul documents are the piece that makes each expert configurable by its owner. The briefing-first home is the consumption side of the same design — the user opens to what the agents did overnight. Release cadence is roughly weekly and the contributor list is small and consistent.
◆Prediction
Given scheduling, credit guardrails and a marketplace now coexist, per-expert monetisation or publishing by outside authors is the obvious next step. The Soul document format is also likely to grow structure.

◆Recent moves

  1. 6d ago

    Expert scheduling, Soul documents, and a briefing-first home

    v0.7.1 fills in the expert model introduced one release earlier: scheduling with attribution, triggers, thread posts and a credit guardrail, editable Soul documents, grouped expert chats, and a briefing-first home. Tavily blocks and Claude Sonnet 5 support arrive alongside. Substantial, but it extends the expert and proactive-output directions already established rather than opening a new one.

    View source ↗
  2. 13d ago

    Rolling synthetic seed fixture for preview databases

    A rolling synthetic seed fixture for preview databases, so PR migrations run against populated tables. Pure CI infrastructure with no user-visible surface; it appears in the feed only because it is published as a repository release.

    View source ↗
  3. 13d ago

    Experts marketplace, scoped sessions, and a Better Auth migration

    ⚡ SPARK

    v0.7.0 introduces expert-scoped sessions with identity context and an experts marketplace with team pages and per-expert threads, while replacing Supabase Auth with Better Auth. It converts the Copilot from one assistant into a roster of scoped specialists, and it does so on a rebuilt identity foundation.

    View source ↗
  4. 20d ago

    Configurable transcription, clipboard images, and Library sorting

    v0.6.70 adds configurable transcription endpoints, clipboard image paste in Copilot chat, most-recently-ran sorting in the Library, and surfacing of skipped webhook presets. Input-surface and housekeeping work that fills in the Copilot experience ahead of the expert split.

    View source ↗
  5. 27d ago

    Agents start posting into Slack and Telegram on their own

    ⚡ SPARK

    v0.6.69 brings proactive posting and DM delivery for Slack and Telegram, plus an agent-building mode with a compaction-proof guide and automatic engine switching. It moves agents from waiting to be asked toward initiating contact in the channels users already occupy.

    View source ↗
  6. 1mo ago

    Maintenance release: tour polish and webhook preset guards

    v0.6.68 is a light release: a README refresh, product tour polish, webhook preset migration guards, and an OpenAPI schema regeneration. Maintenance between the larger platform releases on either side of it.

    View source ↗