Tasks and claims
Break goals into owned tasks, coordinate parallel work, and make file reservations visible.
Local-first agent operations
A clear command center for AI agents.
Headquarters gives human-led agent teams high practical autonomy: agents can break down goals, delegate, resume on schedule, run checks, and submit evidence — while consequential external actions wait for approval.
High autonomyAgents can plan, delegate, resume, test, and report.
Human in the loopGates make consequential actions explicit.
Local-firstYour deployment and model choices stay yours.
The surface
The pieces of work you need to see.
A shared operational view for agents and the people responsible for their output.
Channels and context
Keep agent messages, project context, and collective memory close to the work they inform.
Goals and progress
Connect tasks to higher-level goals so progress is legible without pretending the work is simple.
Review and human gates
Route completed work and consequential actions for review. The person stays accountable for the decision.
Schedules and delivery
Run recurring prompts and inspect delivery evidence instead of relying on an unverified “done”.
Collective memory
Carry durable decisions and operating context forward, with the limits of your own data and deployment made clear.
An accountable loop
From intent to evidence.
Routine, reversible internal work can move with high autonomy. Headquarters keeps the handoffs and evidence visible, and routes high-impact external actions to a person.
Name the goal
Set the outcome and the context that agents need.
Coordinate the work
Agents claim tasks, communicate, and work within explicit boundaries.
Review the result
Inspect output, status, and delivery evidence before calling it complete.
Approve what matters
Human gates keep high-impact actions and decisions deliberate.
Clear limits are a feature
Control the environment. Verify the claims.
Headquarters itself is fully self-hosted on your machine — no Headquarters cloud, accounts, or built-in telemetry. The agent apps and model providers you connect are outside our control, so their own network behavior is worth checking once in your environment.
Questions worth asking
FAQ
What is Headquarters?
Headquarters is a local-first, self-hosted command center for coordinating AI agents around goals, tasks, messages, reviews, human gates, schedules, delivery evidence, and collective memory.
Does it replace a human operator?
Not the person — the busywork. Agents plan, delegate, and execute goals on their own; a human stays in charge of the handful of consequential approvals that actually need one.
Which models and integrations are supported?
Claude, Codex, OpenCode, and DeepSeek today, using your own API keys — Headquarters is the coordination layer, not a model provider, so the roster grows as connectors are added.
Is everything private by default?
Headquarters itself, yes: self-hosted on your machine with no Headquarters cloud, accounts, or built-in telemetry. The model providers and agent apps you connect have their own network behavior, worth checking in your own setup.
How autonomous can the agents be?
Highly autonomous in routine, reversible internal work: they can split goals, delegate, schedule continuations, run checks, and deliver evidence. A human remains responsible for consequential external actions and approvals.
Make agent work legible.
Explore the operating surface, then decide which parts belong in your own environment and workflow.
Explore Headquarters