AutoGPT Platform beta v0.6.62 lands with a focused set of changes that matter most if you are building on top of Copilot or running bots across platforms. The full release is on GitHub.
Copilot gets a skills layer and native scheduling. The headlining addition is a self-distilled skills registry with index injection (PR #13195). Copilot can now build up a reusable index of skills rather than reconstructing context from scratch each session. Pair that with native scheduling for turn followups (PR #13190), and you have the foundation for agents that act on a timeline without external cron glue.
Bot commands are now first-class. Four new commands ship in this release: /new starts a fresh conversation, /resume lets users pick up a past chat, and the bot now posts an intro message automatically when added to a server. A dedicated Bots settings page (PR #13222) adds per-platform link management, so you can wire the same Copilot to multiple platforms without hunting through config files. Server names also stay fresh via gateway events (PR #13221), which removes a common source of stale metadata bugs.
Sharing and visibility improvements. Builders can now share agent chat results via a public link (PR #13081). The Copilot chat UI now shows the source platform for each message (PR #13175), which is useful when you are routing conversations from multiple channels into one dashboard.
Cost visibility gets sharper. A cost breakdown now appears in the agent briefing panel (PR #13129), and a follow-up PR (#13196) cleaned up the UX around it. If you are tracking spend per run, this gives you a native view without exporting logs.
A guardrail for agent creation. Before calling create_agent, the platform now requires a library similarity check (PR #13080). This nudges builders toward reusing existing agents rather than spawning duplicates, which matters for keeping a large workspace manageable.
Deprecated model cleanup. Retired LLM models are being phased out with a family-aware migration path (PR #13089). If your workflows reference older model identifiers, check whether they fall into the deprecated set and migrate before the identifiers stop resolving.
Accessibility fix worth noting. Input placeholder contrast now meets WCAG AA standards (PR #12849). Small change, but it closes a real gap if your agents surface any UI to end users.
What should you do today? If you run Copilot-based agents, enable the skills registry and review the new scheduling primitives. They reduce the scaffolding you would otherwise write yourself. If you manage multiple bots, open the new Bots settings page and consolidate your per-platform links there. And audit any create_agent calls in your automation pipelines since the similarity check is now a required step, not optional.