SupplyMindV3 · Forecast Intelligence
build 3.0.0-rc
BFF probing…·Grok-self-hosted
--:--:-- UTC+2
Agent · Autonomous builder

Agent Studio

A LangGraph agent answers a data question by planning, calling MCP tools, and composing a dashboard from host-approved components. A second model — the judge — verifies it against a rubric before it ships. Semi-deterministic: the agent never writes executable code.

simulated · seedscripted preview · simulation grok-self-hosted (target)
Scripted preview · simulation. This walkthrough replays a fixed script of the planned agent loop — the plan, tool calls, board, and judge verdict are pre-authored, not produced by a live model or a real judge. Run the agent to execute the live LangGraph + AG-UI loop.
Ask
Build trace
1
Plan
Decompose the question
2
Tool calls
Gather grounded evidence
3
Compose
Assemble approved components
4
Judge
Verify against rubric
5
Ship
Render to operator
Generated evidence boardopenui · 5 components
No board yet

Run the agent to watch it plan, call MCP tools, compose an evidence board, and pass it through the judge — live.