Version 3.8.26

🔔 Changelog

📅 October 10, 2026

Introducing AI agents: multi-step assistants that plan, call tools and models, and keep working until they have an answer, unlike the AI chat completion node, which makes a single model call. Each agent has instructions, a default and a fallback model, and an execution budget that limits model calls, tool calls, tokens, duration and retries.
Added the AI Agent node for flows. The flow waits until the agent finishes and continues on Completed, Budget exceeded, Timed out or Fail. You choose which tools the agent may use, whether to continue the conversation's agent session or start a new one, and whether intermediate messages are sent to the customer or kept as internal whispers.
AI agents use your custom functions, MCP clients and AI scopes as tools, with your existing permissions and model billing. An agent fee is billed in addition to model usage.
Flows can now be started when an AI agent run finishes, and agents can also be run from custom functions and the API.
Besides MonoChat Native, AI agents can run on the OpenAI Agents SDK, the Claude Agent SDK, pi or your own endpoint, where available.