// mission control for coding agents

Run a fleet of coding
agents remotely from
one control room.

Operator keeps your agent sessions running server-side, drives the whole fleet in parallel, and tells you exactly which project needs you next.

start at your desk · check from your phone · merge from anywhere
fleet / Driftwatch⌕ Jump to project, session, or command… ⌘K2 NEED YOUA
PROJECTS
+ New project
Driftwatch+ Task
NEEDS YOUR INPUT 1
IN PROGRESS 1
MERGED 1
Driftwatch / task
Tighten incident alerting before launch
1.2M tok · $4.31✦ Fable 5⚑ HighNEEDS YOU
Reviewed the alerting pipeline and thresholds.▌
- alertAfter: 5 * MINUTE
+ alertAfter: 1 * MINUTE, escalate: true
Message the agent…send ↑
↳ interactive replica of the real app · click projects, tasks, and the merge button · mocked data
THE LOOP →
PLAN THE WORK →CREATE TASKS →DELEGATE TO AGENTS →BUILD IN PARALLEL →REVIEW CODE →MERGE →REPEAT →PLAN THE WORK →CREATE TASKS →DELEGATE TO AGENTS →BUILD IN PARALLEL →REVIEW CODE →MERGE →REPEAT →
// 01 · THE PROBLEM

Agents can write the code.
Orchestrating them is still on you.

Sessions die when the laptop sleeps. Tabs pile up. You lose track of which diff is safe, which task stalled, and who’s waiting on an answer. Operator turns the sprawl into one calm queue.

FLEET VIEW

Every project and agent session on one screen.

True parallel sessions running server-side. They keep going after you close the laptop, and a single cross-project signal tells you who needs you.

Sessions persist on the server, not your tab
One rail across every repo and worktree
Cross-project notification when agents finish
PROJECTS · ONE RAIL2 NEED YOU
D1Driftwatch12 tasks · status page & alerts2m ago
M1Menubird3 tasks · menus & online ordering30m ago
PPaperkite1 task · docs → changelog3h ago
LLedgerly1 task · invoices & charts2d ago
HHexbot3 tasks · discord support bot6d ago
↳ all sessions running server-side · safe to close this tab
TASK BOARD

A kanban board where every card is an agent.

Define the tasks for each project, then start them in parallel: every card launches its own agent in its own worktree. Queue what’s next, watch what’s running, merge what’s done.

NEEDS YOU
RUNNING
MERGED
TASK BOARD · DRIFTWATCH4 TASKS · 3 AGENTS LIVE
NEEDS YOUR INPUT 1
Tighten incident alerting before launchHIGHOne diff waiting for your approvalwaiting on you · 7m ago1 session
IN PROGRESS 2
Add uptime sparklines to the dashboardHIGH90-day history per monitor, hover for incidentsagent working · 2m ago1 session
Public API for monitor statusHIGHREST + webhooks, keyed per workspaceagent working · 9m ago1 session
MERGED 1
Dark mode for the status pageHIGHmerged · 1h ago1 session
↳ every card is a separate agent · its own worktree · started in parallel
CONTEXT RECORDER

Task memory survives window resets.

Clearing context condenses the transcript into a summary and seeds a fresh agent with task lineage preserved.

Flight recorder carrying condensed session context between two terminal windowsCTX CARRIED →
SAFE MERGE LANE

Isolated worktree. Review, then merge.

Every task runs on its own branch. Review the diff, then merge it cleanly into main, from your desk or your phone.

Rows of separate sealed chambers, each running one isolated task✓ ISOLATED WORKTREE
INSIGHTS

Know what your fleet costs, and what it ships.

A built-in usage dashboard turns every session into numbers: API-equivalent spend, tokens, tasks shipped and lines merged, over 7, 30 or 90 days.

Daily spend and token trends, cache reads included
Tasks shipped and lines merged, per project
Filter by project, agent, or time range
INSIGHTS · LAST 30 DAYSALL PROJECTS
SPEND▲ 38%$486.20API-equivalent cost
TOKENS USED▲ 24%291.4Macross all categories
TASKS SHIPPED▲ 12%41merged to base branch
LINES MERGED+12.4k-890added / removed on base
DAILY SPENDAPI-EQUIVALENT, PER DAY
↳ every task and token accounted for · per project, per agent, per day
// WHAT'S INSIDE

Everything you need to run a fleet of agents.

01
ALWAYS ON
Agents run server-side and keep working after you close the laptop
02
PARALLEL PROJECTS
Every project builds side by side, one screen across all of them
03
REVIEW & MERGE
Read the diff, merge clean. Conflicts get an agentic fix
04
TASK BOARD
Queue the work. Each task launches an agent in its own worktree
05
CONTEXT CARRY
Clear at high context. A summary seeds the next window, lineage intact
06
PLAN & CURATE
Scope projects, curate the queue, keep the fleet pointed at what matters
// OPEN SOURCE

Run it yourself, or let us run it.

Operator is open source under Apache-2.0. The difference is where it runs: locally on your own machine for free, or hosted by us, always-on and reachable from any device.

SELF-HOST · APACHE-2.0

Local. Free, on your machine.

Clone the repo and run the full app on your own hardware, with the Claude Pro/Max login you already have.

The complete app, nothing held back
Repos, worktrees and transcripts stay on your machine
Runs while your machine is awake
Self-host from GitHub
↳ free · apache-2.0 · your hardware, your login
HOSTED · WAITLIST

Remote. Always-on, any device.

The same app on a dedicated instance we run for you. Start work at your desk, steer the fleet from your phone, merge from anywhere.

Agents keep working after you close the laptop
Reachable from any device, nothing to install
Zero setup: we provision, secure and update it
Get early access
↳ coming soon · join the waitlist

Build at your desk.
Check from the train.
Merge from your phone.

The focused command center for building with agents.

Early access is the hosted, remote Operator: your fleet, reachable from any device. Want it local instead? Self-host free from GitHub.