// the find
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.
AutoGPT has pivoted from the 2023 viral autonomous agent into a visual workflow platform where you wire together blocks (each block = one action) to build multi-step AI agents. The classic recursive agent is still in the repo under `/classic` but is effectively legacy. This is now a competitor to n8n and Langflow, not a standalone AI assistant.
- The block model is a reasonable abstraction — each block has typed inputs/outputs and the graph executor handles retries and parallelism, which is more principled than most agent frameworks that treat 'tool calling in a loop' as an architecture.
- agbenchmark is genuinely useful: it runs objective, automated evaluations against agent protocol-compatible agents, so you can actually measure whether a change made your agent better or worse instead of vibes-testing it.
- MCP (Model Context Protocol) support is already wired in, meaning external tools can integrate without writing bespoke blocks.
- Multi-model support covers OpenAI, Claude, and Llama API, so you're not locked to one provider at the block execution level.
- The `autogpt_platform` folder — the actual current product — is under Polyform Shield, not MIT. You cannot build a competing service on it. If you're evaluating this for anything commercial, read that license first, because the README buries it.
- Self-hosting requires Docker Compose, 8GB RAM, Node, npm, and several running services. The one-line install script pulls from `setup.agpt.co` with no checksum verification, which is a supply chain risk you'd normally reject in a code review.
- The cloud-hosted version is still in closed beta with a waitlist, so the 'just use the hosted version' escape hatch doesn't exist yet. You're either self-hosting this complexity or waiting.
- The repo carries three years of pivots: classic agent, Forge, benchmark, new platform — all coexisting. Navigating what's current vs. what's abandoned is genuinely confusing, and the 185k stars are almost entirely from the 2023 hype cycle, not from the current platform.