A self-hosted dashboard that treats AI agents like production infra

What it does

Mission Control is a self-hosted Next.js dashboard for managing AI agent fleets. It wraps task dispatch, multi-agent workflows, token-cost tracking, and security auditing into a single SQLite-backed application that runs without Redis, Postgres, or Docker. You register agents, move tasks across a Kanban board, and monitor trust scores and memory graphs from one interface.

The interesting bit

Most open-source agent tools stop at prompt chaining; this one treats operations as a first-class concern. It includes a four-layer evaluation framework, mandatory quality gates that block task completion until signed off, and hook profiles ranging from minimal to strict. The gateway-optional mode is a nice touch: you can use the task board, scheduler, and cost panels even if your agents are just local CLI scripts with no gateway connection.

Key highlights

• 32-panel SPA covering tasks, agents, skills, memory, cron, alerts, webhooks, and cost tracking.
• SQLite-only core with no mandatory external dependencies; hardened Docker profiles are available but optional.
• Multi-gateway adapters for OpenClaw, CrewAI, LangGraph, AutoGen, and Claude SDK.
• Security tooling includes RBAC, secret detection, MCP call auditing, and per-agent trust scoring.
• Natural-language cron scheduling and a four-layer eval framework with drift detection against a 4-week baseline.

Caveats

• Explicitly alpha: APIs, database schemas, and configuration formats may change between releases.
• Several feature screenshots were temporarily removed from the README because they no longer reflect the current UI.
• The authors advise reviewing security considerations before deploying to production.

Verdict

Worth a look if you are running multiple agents and need centralized observability with cost and security controls. Skip it if you want a stable, battle-tested platform or only run a single agent sporadically.