Recruiter walkthrough of a self-hostable multi-agent operations platform.
These are real screenshots from a running instance — streaming chat,
DAG missions with human-approval gates, RAG, MCP, evals, and cost budgets.
This page is a static gallery, not a hosted FastAPI + Next.js deploy
(that stack needs Postgres).
Landing — pillars plus the ops layer, captured from a
running instance. The product is open source and self-hostable.
This gallery is not a live app URL.
Chat — streaming multi-agent chat with a roster
switcher. This capture is the empty conversation shell and suggested
prompts, not a live tool-call. Tool-call cards appear when an agent
actually calls a tool (web search, document retrieval, or MCP).
Missions — the orchestrator turns a goal into a live
task DAG. This capture is a competitor-teardown run paused on a
human-approval gate (“Post the digest to Slack”). Token counts on the
cards are from that run, not product SLOs.
Evals — Eval Center for the Clinical Research agent:
golden cases plus a recorded run. This capture shows 3/4 passed on
that one run — not a fleet-wide score. CI runs deterministic
checks; the LLM-as-judge path is documented in the README. No invented
eval metrics.
