// the find
builderz-labs/mission-control
Self-hosted AI agent orchestration platform: dispatch tasks, run multi-agent workflows, monitor spend, and govern operations from one mission control dashboard.
Mission Control is a self-hosted Next.js dashboard for running AI agent fleets: dispatch tasks, coordinate multi-agent workflows, track token spend, and audit security events. It targets teams running Claude Code, CrewAI, LangGraph, or AutoGen who want a single pane of glass instead of scattered logs and curl calls. It's alpha, explicitly warns of schema churn, and is clearly still finding its scope.
577 tests (282 Vitest unit + 295 Playwright E2E) is unusually thorough for a dashboard project at this star count — most projects this age have near-zero E2E coverage. Zero hard external dependencies: SQLite in WAL mode, single pnpm start, no Redis, no Postgres required — that makes self-hosting on a $5 VPS realistic. The security surface is concrete rather than vague: scrypt password hashing, HMAC-SHA256 webhook signatures, per-agent trust scoring, MCP call auditing, and secret detection on agent messages are all real features, not marketing bullets. OpenAPI 3.1 spec with 101 endpoints and a Scalar UI ships with the repo, which means the API is actually documented and testable without guesswork.
SQLite is fine until it isn't — WAL mode handles concurrent reads but serializes writes, and there is no migration path documented for when a busy agent fleet saturates it. The project is 39 migrations deep and self-describes as alpha with breaking schema changes expected, so adopting it now means either freezing at a version or absorbing upstream churn. The scope is dangerously wide: office sprite animations, i18n in 10 languages, a built-in TUI, PTY terminal, MCP server script, 'SOUL personalities' for agents, and a pixel-art office sim coexist with the serious orchestration features — this is a sign the project is still figuring out what it is. The real-time layer (live session updates, agent messaging) depends on 'OpenClaw', which appears to be the team's own proprietary gateway; without it you're in polling mode and lose the features that make the dashboard actually useful at scale.