Overview
Use these guides when you already understand the main Flow AI concepts and want to complete a specific implementation task.
Use these guides when you already understand the main Flow AI concepts and want to complete a specific implementation task.
If you are new to Flow AI, start with the Quickstart first.
Run and inspect an agent locally
Studio
Run a local browser UI, chat with agents, inspect traces, create tests, and run evals.
Streaming events
Consume runtime events from your application or HTTP layer.
Connect agents to data
Data environment
Configure catalog, search, KV, target database, and knowledge dependencies.
Catalog profiling
Profile a database and export a catalog artifact.
Knowledge and documents
Index source material and expose retrieval-backed context to agents.
References and glimpses
Pass large, sensitive, or intermediate payloads through agent workflows by pointer, while giving agents compact metadata for planning.
Control execution
Require approvals
Pause plans or tools before sensitive work continues.
Execute approved actions
Apply approved plan actions to your platform.
Open Execute approved actions.
MCP
Expose Flow AI tools to MCP-aware clients.
Test and evaluate
Test agents without provider calls
Run deterministic and scripted interpreters without live model calls.
Open Test agents without provider calls.
Evals
Score trajectories, actions, and runtime-backed flows.
Final-response judge evals
Score response quality with rubric-driven model judges.
Open Final-response judge evals.
Troubleshoot
Debug system prompts
Inspect where prompt text is defined, rendered, passed into the runtime, and sent.