Self-hosted control plane for operating AI agents.
Dispatch tasks, inspect runs, review failures, track spend, and coordinate agent runtimes
from one local dashboard backed by SQLite.
An open-source project by Builderz Labs, created and maintained by nyk.
It works with OpenClaw, Claude Code, Codex, and other runtimes - it is not part of any one of them.
[!WARNING]
Mission Control is alpha software. APIs, schemas, and configuration may change between
releases. Read the security guidance before exposing it to a network.
Start locally
Node.js 22 or newer and pnpm are required for a source install.
git clone https://github.com/builderz-labs/mission-control.git
cd mission-control
bash install.sh --local
Open http://localhost:3000/setup, create the first admin account, then copy the API key
from Settings if an agent or script needs headless access.
The manual path is useful when you already manage Node and pnpm:
nvm use 22
pnpm install
pnpm dev
Windows users can run ./install.ps1 -Mode local in PowerShell.
Start with Docker
docker compose up
Or run the published multi-architecture image:
docker pull ghcr.io/builderz-labs/mission-control:latest
docker run --rm -p 3000:3000 ghcr.io/builderz-labs/mission-control:latest
Use the hardened Compose overlay for a network-accessible deployment:
docker compose -f docker-compose.yml -f docker-compose.hardened.yml up -d
The deployment guide covers persistent data, TLS termination,
gateway connectivity, and standalone builds.
What Mission Control governs
The control plane sits above agent runtimes. It does not replace their reasoning or tool
loops. It gives operators one place to see and govern the work around those loops.
Area
Shipped surface
Tasks
Inbox, assignment, execution, review, Aegis quality gate, and completion receipts
Agents
Registration, presence, sessions, runtime adapters, configuration, and workspace sync
Memory browser, relationship graph, skills registry, and local skill synchronization
Governance
Roles, API keys, security events, trust signals, approvals, audits, and evals
Interfaces
Web UI, CLI, MCP server, OpenAPI-described REST API, WebSocket, and SSE
The runtime is self-hosted and workspace-aware. SQLite stores local control-plane state.
Shared workspaces can use deployment-level runtime integrations. Strict workspaces block
those integrations until the underlying resources carry workspace ownership. A gateway is
optional for task, project, agent, scheduler, webhook, alert, and cost work; live session
messaging needs a connected runtime gateway.
Operator field notes
Dashboards compress a sequence into current state. When a run needs review, record the
identity, task, tool call, approval, result, and verification evidence before changing it.
Keep unresolved items distinct from accepted risk.
Logs show what ran. A completion receipt or inspected artifact shows what finished.
Pick the right fit
Use Mission Control when multiple agents or runtimes make it hard to answer who owns a
task, what executed, which result passed review, or where spend and failures accumulated.
It is probably the wrong tool when:
one agent on one machine already stays understandable from its native CLI;
you need a managed multi-tenant SaaS rather than a self-hosted control plane;
you want an agent framework to define planning and tool use;
your deployment cannot tolerate alpha schema or API changes.
Adapters and observation surfaces cover OpenClaw, Claude Code, Codex,
CrewAI, LangGraph, AutoGen, and Claude SDK workflows. Adapter depth varies by runtime; see
agent setup and CLI integration before
assuming feature parity.
Connect an agent
The shortest gateway-free loop uses the REST API. Export the URL and API key shown in
Settings:
Use the CLI and MCP reference for the current command and tool
surface. The REST contract lives in openapi.json. A running instance serves
the interactive reference at /docs and the OpenAPI JSON at /api/docs.
Product surfaces
Tasks and quality review
The task board tracks work through inbox, assignment, execution, review, quality review,
and completion. Aegis review requires an approval record before a task reaches done.
Agents and runtimes
Agent views combine registration state, heartbeats, sessions, configuration, local runtime
discovery, and workspace files.
Memory and skills
The memory browser and relationship graph inspect filesystem-backed memory and linked
session knowledge. The Skills Hub discovers local skill roots and scans registry content
before installation.
Schedules and activity
Recurring task templates create dated work on a cron schedule. The activity stream combines
agent, task, and system events for operator review.
Session cookies, API keys, Google sign-in, and role checks
Validation
Zod at input boundaries
Verification
Vitest, Playwright, ESLint, TypeScript, build, and API parity checks
Runtime data defaults to .data/. Set MISSION_CONTROL_DATA_DIR to an absolute persistent
path for standalone deployments. The complete environment contract is in
.env.example.
Security boundary
Keep Mission Control on a trusted network unless a TLS reverse proxy and
MC_ALLOWED_HOSTS are configured.
Replace or securely store generated credentials before broader access.
Use the hardened Compose overlay for production-like container deployments.
Treat agent messages, skill packages, webhooks, and MCP content as untrusted input.
Report vulnerabilities through SECURITY.md, not a public issue.
Access controls and security inspection surfaces are included, but alpha status
still applies. Read SECURITY-HARDENING.md before relying on a
network-accessible deployment.
Mission Control 是一个自托管的控制平面，用于管理 AI 编程代理。它可以分派任务、审查运行情况、跟踪开销，并通过一个基于 SQLite 的本地仪表盘，统一协调 OpenClaw、Claude Code 和 Codex 等运行时。