AutoGPT Platform Beta v0.6.63 adds local LLM support and global search

Significant-Gravitas has released AutoGPT Platform Beta v0.6.63, a major update to the open-source autonomous AI agent platform (185k stars on GitHub). The release introduces a local-LLM AutoPilot mode, allowing users to run agents entirely offline without an API key. This is a game-changer for privacy-conscious users or those with limited internet access. Additionally, the platform now features a global search endpoint with a bucketed command-palette modal, enabling quick navigation across the entire UI. Another standout is the auto-discovery of MCP (Model Context Protocol) integrations when no native block exists, streamlining third-party tool connections.

The update also includes goal decomposition: the copilot now breaks down complex goals into steps before building an agent. New Artifacts page (behind a feature flag) provides richer artifact cards with optimized file preview. Blocks like Replicate now support file inputs, and the Execute Code block gets variable inputs. On the business side, a bidirectional Stripe tier reconciliation with provenance and periodic sweep improves subscription management, along with a subscription pricing experiment in the frontend. Numerous fixes address OAuth errors, Discord bot threading, and frontend input handling. The release also mentions DataFast revenue attribution on Stripe Checkout.