AutoGPT Review

AutoGPT Platform has four interfaces. AutoPilot turns a plain-English job description into a working agent, the Visual Builder is a canvas of blocks for branching and looping, the Dashboard shows agent status, run history, spend, and items needing attention, and the Marketplace offers community agents you can add and customize. Agents run on demand, on a schedule, or from a trigger, and integrate with 45+ platforms such as Gmail, Slack, Google Sheets, Notion, HubSpot, GitHub, and Airtable, with access to hundreds of AI models. Blocks are Python classes with an input schema, an output schema, and a run method, which makes the system extensible in code.

The README calls self-hosting the free path, but you provide infrastructure, model API keys, and maintenance, and support is community-only. The planned appliance installer is a single container with one loopback port and needs a running Docker daemon with Linux containers on amd64 or arm64; the hosted installer is not live yet, the first release targets Linux and macOS, and Windows users should follow the from-source setup. The manual path is to clone the repository, run make init-env in autogpt_platform (it generates secrets, and the backend refuses to start while ENCRYPTION_KEY is empty), and run docker compose up -d --build. The frontend is on port 3000, the backend WebSocket on 8001, and the execution REST API on 8006. Core services are Postgres with pgvector, Redis, and RabbitMQ, and authentication now uses a built-in Better Auth service instead of the older bundled Supabase stack. The docs we read give no CPU or RAM requirements.

For non-programmers, AutoPilot and the Marketplace are the entry points, and for builders the Visual Builder exposes each step as a block. The Dashboard emphasizes cost and run history, which matters because agents consume model credits. This section reflects the repository and documentation; we did not run the platform.

• AutoPilot: build an agent from a plain-English description
• Visual Builder with blocks, branching, and looping
• Dashboard for status, run history, spend, and alerts
• Marketplace of ready-made community agents
• Run on demand, on a schedule, or from a trigger
• 45+ integrations and hundreds of AI models
• Custom blocks written as Python classes with typed schemas

• Both a no-code route (AutoPilot) and a code route (custom blocks) for the same agents
• Cost and run history are first-class in the dashboard
• Very large community and brand recognition, with about 188,000 stars
• A hosted option with built-in model access for people who do not want to run infrastructure

• The platform's Polyform Shield license is source available, free for personal and internal business use but not for selling a competing hosted service
• The single-container installer is gated on release criteria and not available yet, and Windows needs the manual setup
• Self-hosting needs Postgres, Redis, RabbitMQ, and secrets management, with community-only support
• Autonomous agent reliability depends on the model and the task, and runs consume credits
• Hosted pricing and execution limits are not documented on the pages we read

Self-hosting is free, and you pay for your own model API usage and infrastructure. The hosted AutoGPT Platform is paid, with usage-based agent runs and built-in model access so you do not need API keys, and the pages we reviewed list no prices. The classic code in the repository is MIT-licensed.

Claude Managed Agents is Anthropic's hosted infrastructure for running Claude agents, while Relevance AI, Lindy, and Microsoft Copilot Studio are commercial agent builders. AutoGPT is the closest widely used builder you can self-host, with a visual canvas, a marketplace, and a block system, at the price of a source-available license on the platform. Choose it when you want agents on your own infrastructure with your own model keys, and read the Polyform Shield terms if you plan to offer it to others.

• People who want to build recurring agents such as briefings, research, and monitoring with a visual tool
• Teams that want to self-host an agent platform with their own model keys
• Developers who want to extend a platform with typed Python blocks
• Users who want to start from marketplace agents

• Anyone who requires an OSI-approved open-source license for the platform
• Teams that need guaranteed reliability for critical, unattended workflows
• Windows users who want a one-click self-hosted install today