AutoGPT

AutoGPT started as one of the first "give it a goal and let it run" autonomous agent scripts. It has since grown into the AutoGPT Platform: a full environment for designing, running, and sharing AI agents rather than a single command-line tool.

The current release is autogpt-platform-beta-v0.6.65, published on June 25, 2026. The recent v0.6.6x line (May-June 2026) centers on AutoPilot/Copilot: automatic goal decomposition, an AutoPilot context panel with shell and Files tabs, a global Cmd+K search, MCP auto-discovery, public link sharing of agent chat results, and native scheduling and webhook triggers.

Capability-wise you get a drag-and-drop Agent Builder, an agent marketplace/store to publish and reuse agents, and 30+ integrations including Slack, Discord, and Stripe. Agent runs are metered through credit-based execution billing, and you can self-host or use the managed AutoGPT Cloud.

Pricing is flexible. The self-hosted open-source core is MIT-licensed and free ($0) if you bring your own LLM API keys and compute, with no seat limits. Managed cloud adds Pro at $50/mo ($42.50/mo billed annually, $510/yr) and Max at $320/mo ($272/mo billed annually, $3,264/yr); a Team tier is coming soon. A separate pay-as-you-go credit wallet meters agent runs at an unpublished rate.

Limitations are worth noting. The platform is still officially beta, autonomous agents can burn credits or API tokens unpredictably, and the pay-as-you-go run pricing is not published up front. Self-hosting removes the subscription cost but shifts setup, LLM keys, and maintenance onto you.

AutoGPT suits developers and technical teams who want to build multi-step autonomous agents visually, publish or reuse agents from a store, and choose between free self-hosting and a managed cloud with predictable subscriptions.