Agent Playground: Phase 1 + 2

Routing, tool calling, traces, and replay.

Defaults use a simulated provider (`mock://`) so you can explore the flow without running Ollama or any external model.

Inputs

Docs JSONAgents JSONPatch JSONUser Input

Router Panel

Auto-routeRouter modelRouter endpointManual override agent

{}

Selected: n/a

Prompt Panel

Main modelMain endpoint

No run yet

[]

Tool approvals

Allow fs.writeAllow shell.execAllow http.requestHTTP allowlist domains (comma-separated)

Timeline Panel

Simulate tool call

{}

Run History + Replay

What happened?

Each run can route input to an agent, resolve prompt variables, execute tool calls under approval gates, and emit a full trace timeline for debugging.