Overview
Build AI agents that act autonomously — tool calling, MCP integration, multi-step workflows, and human-in-the-loop patterns.
Agents that act
Agentic integration builds AI agents that act autonomously: they decide what to call, in what order, and whether to hand back to a human. bionicjs provides thin starting points on top of the LLM provider you chose.
Tool calling
Tools are typed functions your agent can invoke. Declare them next to your model configuration, and the framework generates the callable surface your agent uses to act on your app's data.
MCP
Model Context Protocol servers connect your app's tools and data to agent clients. Enabling MCP gives you a deterministic server exposing the same capabilities your tools define.
Multi-step workflows
Agents can run several steps and re-plan between them. Each step is an ordinary tool call, so the whole workflow is inspectable, resumable, and testable like any other code path.
Human-in-the-loop
Long-running or risky actions hand control back to a person before they complete. The agent pauses on a defined boundary and resumes only when the human approves.