Agent observability

Stop guessing what your agents did.

Traces, token cost, and tool calls for every session. Locally, in the IDE.

› Spotlight: analyze and visualize with agents — one-click sharing with your team

Reimagined for agents: ask your data analyst agent, plot charts, build dashboards — all locally — and share them with your team.

› Agent packs: share work in the agentic era

Built to speed up innovation. Fully reproducible, with one-click publish and deploy the entire agent.

Fastest tool to build agentic software: author agent packs and skills, exercise them on any agent engine in the UI or CLI, then hand off — teammates install in one click from your library or the store.

Store catalog pulls curated agent packs your org can adopt without rebuilding the same glue.

› The pain you know

Invisible loops

Your agent retried 11 times before you noticed. Traces show each step.

Surprise bills

Token cost per session, per step. Know before the invoice arrives.

Black-box agents

Tool calls, prompt sizes, model choices. Visible, not guessed.

› What you get

Session traces

Every tool call, every model response, every token. Locally stored via OTEL.

Spotlight analytics

Ask for charts in plain language. DuckDB queries on your trace data.

Multi-provider

Claude, Codex, ChatGPT, Cursor, LiteLLM. Same traces format across all.

Local-first

Traces stay on your machine. Your data, your storage, no third-party pipeline.

› When seeing the spend isn't enough

Visibility tells you the loop burned $300. Control stops it at $5. In governed deployments, every agent carries a hard budget enforced before the model call — the request past the cap is refused, and every dollar attributes to the user who asked. That's the agent control plane: the same trace data you debug with, promoted to enforcement.

Download free See plans