← All reportsPublic report

significant-gravitas/autogpt

AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.

187k stars Python View on GitHubprofiled 11d ago
Try GitZoid
148PRs this week
17Contributors
0Deps scanned
0Issues found
01 · Repo overview

How autogpt is put together

Monorepo for AutoGPT containing two product lines: 'classic' (the original Python autonomous-agent framework with forge/original_autogpt/direct_benchmark packages, Poetry-managed) and 'autogpt_platform' (a Next.js/TypeScript frontend talking to a FastAPI/Python backend whose data model is defined by schema.prisma over PostgreSQL, with Redis and RabbitMQ-style messaging in the dependency set). A GitHub Actions ChatOps 'batch-deploy bot' (described in README.md) merges approved PRs onto ephemeral batch/rollup-* branches, opens a rollup PR, triggers per-PR preview environments via '!deploy' comments, and lands batches via '/batch-merge' with auto-merge squash. A single-container deployment path (Dockerfile + shell bootstrap scripts) packages the whole platform. Data flow: frontend (orval-generated React Query client against backend OpenAPI) -> FastAPI backend -> Postgres (Prisma) with per-preview schema isolation.

Languages

PythonTypeScriptJavaScriptSQL

Frameworks

FastAPINext.jsPrismaTailwind CSSReact Query (orval-generated)PlaywrightPydantic

Datastores

PostgreSQL (Prisma, per-preview schemas)Redis

Infrastructure

Docker / docker-composeGitHub Actions (ChatOps workflows)Vercel (preview URLs *.vercel.app)SentryCloudflare Worker (cloudflare_worker.js)Google Cloud Storage/Logging clients

Major components

autogpt_platform/backend

FastAPI backend service implementing the AutoGPT Platform API, agent execution (app.py, batch_executor.py), and database access (db.py) against the Prisma-defined Postgres schema.

autogpt_platform/frontend

Next.js web UI with an OpenAPI-derived React Query client generated by Orval, Tailwind styling, Sentry instrumentation, and Playwright E2E tests.

autogpt_platform/autogpt_libs

Shared platform library providing JWT-based auth (pyjwt[crypto]), cryptography, Redis, settings, and Google Cloud logging helpers.

classic/

AutoGPT Classic autonomous-agent framework: forge core library, original_autogpt CLI app, and direct_benchmark evaluation harness.

batch-deploy bot (GitHub Actions)

ChatOps workflows (/batch, /batch-remove, /batch-merge) that maintain batch:* labels, rebuild batch/rollup-* branches from dev, manage rollup PRs, and trigger/tear down preview deployments.

autogpt_platform/single-container

All-in-one Docker image and shell entrypoint/bootstrap scripts that run the full platform stack in one container.

autogpt_platform/analytics

Collection of SQL queries for retention, onboarding-funnel, execution, and cost analytics.

autogpt_platform/graph_templates

Prebuilt agent graph JSON templates (Discord bots, Medium blogger) shipped with the platform.

Over the past month the AutoGPT platform has been building out a full 'experts' product — hireable AI team members with their own memory, schedules, work surfaces, and a briefing-first home page — alongside a major authentication migration to Better Auth and new Stripe payment flows. The team also shipped many integrations (Teams bot, Tavily, All Quiet, DataForB2B), hardened security and reliability across backend services, and cleaned up infrastructure like single-container distribution. The final week was a quiet one, limited to a small database routing fix.

Week by week

2026-08-24A quiet week with just one maintenance fix to how read-only database queries are routed.latest1 change

Fix

Safer routing for read-only database queries

Read-only graph database queries now go through a dedicated query channel, keeping them separate from write operations.

2026-08-17A big week for the experts product — hiring, firing, memory management, and expert groups all landed — plus new payment blocks and a Microsoft Teams bot.6 changes

Feature

Hire your own AI experts

Users can now raise (create) their own custom experts, organize them into named groups called 'pods', and give them a work surface with work cards.

Feature

Expert lifecycle management

Added admin tools to view an expert's memory, isolate it per user, and cleanly archive ('fire') experts when they're no longer needed.

Feature

Microsoft Teams bot adapter

The platform can now connect to Microsoft Teams so users can interact through that chat app.

Feature

Stripe Link wallet payments

New device-code sign-in flow plus wallet blocks let users pay via Stripe Link, including support for paying MPP merchants.

Feature

New integrations and subscription triggers

Added All Quiet incident-management blocks, DataForB2B business-data provider, and Stripe subscription webhook trigger blocks.

Feature

User-facing memory settings page

A new settings page lets users control how the platform's memory features behave for them.

2026-08-10The home page became a morning-briefing hub with per-expert spending, while account linking, single-container deployment, and a ChatGPT/Codex transport preview moved forward.6 changes

Feature

Briefing-first home page

A redesigned /home page leads with an AI-narrated morning briefing showing what needs attention and each expert's spending.

Feature

Chat-to-platform account linking

Users can now fully connect their chat accounts to the platform, completing the loop started earlier.

Feature

Single-container distribution

The platform can now be packaged and published as one container image, making self-hosting simpler.

Feature

ChatGPT/Codex subscription transport preview

An early preview lets eligible users use their existing ChatGPT/Codex subscriptions as a way to run the platform's AI.

Feature

Day-one kickoff after hiring an expert

Newly hired experts now start working immediately with a first-day kickoff instead of waiting around.

Feature

Team sidebar with presence

The sidebar now shows team members and whether they're active, with a menu to create new ones.

2026-08-03A 40-commit week centered on voice-driven onboarding, editable expert 'Soul' documents, scheduling, and a sweep of security patches plus new model and search integrations.6 changes

Feature

Voice brain-dump onboarding

New users can onboard by speaking their thoughts aloud, with a quality gate ensuring the transcription is good enough to use.

Feature

Editable expert Soul documents

Each expert now has an editable 'Soul' document defining its personality and behavior.

Feature

Claude Sonnet 5 and Kimi K3 support

The platform added Claude Sonnet 5 (with tokenizer awareness) and consolidated all supported AI models into a single catalog.

Feature

Tavily search integration

New blocks let agents use Tavily for web search, page extraction, crawling, and site mapping.

Fix

Security dependency patches

Vulnerable dependencies were patched across the frontend, backend, and classic AutoGPT, and video downloads gained SSRF protection and size limits.

Feature

Expert scheduling groundwork

Experts can now be scheduled to run tasks, with proper attribution, triggers, and credit guardrails in place.

2026-07-27The headline change was replacing Supabase Auth with Better Auth, alongside the first Expert entity APIs and an experts marketplace section.6 changes

Feature

Authentication system swap

The platform replaced Supabase Auth with Better Auth as its login and identity system.

Feature

Experts marketplace and team pages

A new marketplace section showcases available experts, with team pages and per-expert chat threads.

Feature

Expert entity and hire API

The backend gained the core Expert data model, an API to hire or install experts, and starter rosters of pre-built experts.

Feature

Compact wallet popover

A smaller, tidier credits/wallet display arrived in the new interface layout.

Fix

Memory queue reliability

Memory operations now retry when the FalkorDB queue overflows instead of silently dropping them.

Fix

Clearer upload errors

Image and logo uploads now show explicit file-size error messages instead of failing quietly.

03 · Security check

Dependencies and code review

0 dependencies scanned

Dependency advisories

Security Watch

No known advisories across 0 scanned dependencies.

No known advisories in the scanned dependencies.

Code review

No concrete code-level issues confirmed.

Get this report every week for your repos.

GitZoid learns each repo, reports what changed, and flags what needs attention. One flat price for the whole team.

$19 a month, flat · First 10 outputs free · No card required