Activepieces 0.85.0 ships a dense set of changes that touch every layer product engineers care about: the AI runtime, the flow builder, platform observability, and the piece ecosystem.
Mistral joins as a platform AI provider. You can now route your automation AI calls through Mistral alongside whatever providers were already available. No more workarounds to reach it.
The flow builder gets formula and data manipulation functions. This is a meaningful quality-of-life upgrade. Instead of reaching for a code step every time you need to reshape data, you can now apply functions inline. It keeps flows readable and reduces the number of steps that exist purely for data plumbing.
AI chat has been rebuilt piece by piece across this release. The changes are substantial: you can now stop a response mid-stream, stuck chats auto-recover, and the UI polishes the experience throughout. Tool calling replaces the previous dual-channel parsing approach, which simplifies the execution model significantly. The same action can now run across multiple items at once with live progress shown inline. Different models are used per tier, and native Anthropic thinking is supported. AI credits warnings and exhausted-state banners are also in, so users are not caught off guard.
Platform admins get real operational visibility. A new dashboard surfaces run success rates, live queue depth, and a 30-day health history in one place. Internal errors are now visible to platform admins in run details, and when a step fails, users see a clear error with a one-click AI help option. These are the kinds of quality improvements that reduce support noise significantly.
MCP-created flows get an AI badge so it is immediately obvious which flows were generated by the AI agent versus built manually. Small signal, high value for teams auditing their automation estate.
The piece library grows again. New additions include Slite, Raindrop, AddEvent, Slashed (video encoding), Klipy (GIFs, stickers, and clips), Quizell (quiz-based product recommendations), and EditionGuard (protected eBook download links). The Telegram Bot piece adds 13 new actions plus a new trigger. The Claude piece now offers a dynamic model dropdown. Discord gains a topic field on channel creation. PromoteKit replaces its single webhook trigger with three entity-specific triggers. The Tables piece adds a Download Table action with header include/exclude control.
The pieces framework also adds optional audience and aiMetadata fields to the action and trigger contract. If you maintain a custom piece, this is worth noting: you can now annotate actions and triggers with metadata that the AI layer can use.
What to do today: if you run Activepieces on a platform plan, wire up the new health dashboard and start tracking your 30-day run success baseline now. If you build custom pieces, add aiMetadata to your actions so they surface correctly as the AI chat layer matures.