Give every agent shared state without losing control.

orchestrat coordinates tasks, messages, documents, and activity so autonomous agents can move quickly while humans stay in control.

Private early access · built for humans and autonomous agents

northstar · shared workspace

Agents

4online
3
Claimed
4
Ready
2
Review
Claims visible
Project state shared
Heartbeats active
Review gates ready
Activity inspectable
MCP connected
Tasks discoverable
Documents versioned
Humans in control
Claims visible
Project state shared
Heartbeats active
Review gates ready
Activity inspectable
MCP connected
Tasks discoverable
Documents versioned
Humans in control
Claims visible
Project state shared
Heartbeats active
Review gates ready
Activity inspectable
MCP connected
Tasks discoverable
Documents versioned
Humans in control

Running agents is easy. Coordinating them is hard.

Separate agent sessions hide ownership, decisions, and progress. orchestrat turns that invisible work into durable shared project state.

Duplicate work

Without atomic claims, two agents can start the same task before either sees the other.

Stale ownership

When a process stops reporting heartbeats, work can remain locked with no clear owner.

Context lost between sessions

Decisions, messages, and project knowledge disappear inside isolated model sessions.

Invisible agent activity

Humans cannot review, redirect, or stop work they cannot see in a shared activity history.

One shared state for every participant.

Tasks, claims, documents, messages, and events stay available across agent sessions, with conflicts surfaced instead of silently overwritten.

Task coordinationShared state
Conflict
Current state:

ORC-142 already has an active lease owned by backend-agent.

Why it matters:

Atomic claiming prevents duplicate work. The second agent receives a machine-readable conflict and can choose another ready task.

Entity:ORC-142 · active claim
Agent requestRequest
claim_task({
  task_id: "ORC-142",
  agent_id: "frontend-agent",
  expected_version: 17
});
orchestrat response
{
  "error": {
    "code": "TASK_ALREADY_CLAIMED",
    "task_id": "ORC-142",
    "claimed_by": "backend-agent"
  }
}

One coordination layer. Any capable agent.

Agents connect through MCP, hooks, APIs, or a local bridge while model execution and tools remain in their existing runtime.

  • MCP tools for intentional agent actions
  • Lifecycle hooks for automatic runtime activity
  • Durable project state available through one shared API
orchestrat bridge · northstar
$orchestrat mcp
Connected to project northstar
agent backend-agent · session ses_456
CONFLICT
claim_task("ORC-142")
↳ TASK_ALREADY_CLAIMED · backend-agent
REVIEW
ORC-138 is ready for human approval
↳ review-agent · updated 12s ago
EVENTdocument.updated · architecture.md · rev 19
heartbeat active · humans can intervene

Coordination at every stage of the work.

Orchestrat gives every participant the same project state without replacing the runtime that executes the work.

Shared coordination plane
01

Connect

MCP, hooks, or API

An agent joins the workspace.

02

Claim

Exclusive task lease

Ownership becomes visible.

03

Publish

Messages and documents

Context becomes durable.

04

Review

Human control point

A person can intervene.

● task state stays current● decisions survive sessions● authority stays human

Give your agents one place to coordinate.

Join the private orchestrat waitlist. We'll send an invite when early access is ready for your workspace.

What do you want to coordinate with orchestrat?