Captains orchestrate agent swarms

A Captain is a long-lived agent that splits your work across hundreds of threads, on any machine and any model. It reads every report and tells you only what needs you.

Autumn
Supermemory
Exa
Warp
Pocket
Melius
Infisical
Krea
Goblins
RunPod
Town
Firecrawl
Aside AI
Taste Labs
Revyl
Finta
Autumn
Supermemory
Exa
Warp
Pocket
Melius
Infisical
Krea
Goblins
RunPod
Town
Firecrawl
Aside AI
Taste Labs
Revyl
Finta
Autumn
Supermemory
Exa
Warp
Pocket
Melius
Infisical
Krea
Goblins
RunPod
Town
Firecrawl
Aside AI
Taste Labs
Revyl
Finta
Autumn
Supermemory
Exa
Warp
Pocket
Melius
Infisical
Krea
Goblins
RunPod
Town
Firecrawl
Aside AI
Taste Labs
Revyl
Finta
A Captain in Capy: a one-line merge answered in its chat, beside the items waiting on you and its 120 threads with their models and pull requests
A Captain’s Overview: three items waiting on you above its 120 threads, each with its model and pull requests
Captain capy

One agent runs the swarm

Tell a Captain what it’s for. It starts a thread for each piece of work, hears from each one when it finishes a turn, and keeps the plan current.

lisa9:02 AM

Ship everything in the Q4 roadmap

Captain
  • Readq4-roadmap.md
  • Read120 issues
  • Created120 threads

Splits the work

One ask becomes a thread per piece. Each thread is a full agent with its own machine, branch and pull request.

Reports from threads arriving in a Captain’s chat as each one finishes a turn

Reads every report

Each thread reports when it finishes a turn or needs something. The Captain reads them all and updates its plan.

Capy in a hard hat

Across your machines

Threads run in the cloud by default, or on a computer you’ve paired. Pick for every thread at once, or name a machine for one.

Capy assembling a stack of machines

Cloud machines

Each thread boots its own cloud machine with your repositories checked out, so a hundred threads never share a disk.

The Captain setting for where threads run, open on Cloud, a Mac mini and a build server, with the Mac mini picked

Your own computers

Point every thread at a paired Mac, PC or build server, or say “Run this one on my Mac mini.”

Capy in glasses

Any model, your keys

Threads use the Captain’s model unless you or the Captain pick another, and they run on the subscriptions and keys you connected.

The model picker’s favorites: Claude Fable 5.2, Claude Opus 5.5, Claude Sonnet 5.5, GPT-6.1 Sol and GPT-6 Luna

A model per thread

The strongest model for the hard refactor, a fast one for small fixes. Pick per thread, or let the Captain choose.

Your subscriptions and keys

Every thread runs as you, on your Codex, Copilot or other subscriptions and your own API keys.

Pelican

Steer from anywhere

Message a Captain from the app, from Slack, or from your phone’s browser. It picks up the message wherever you sent it.

  • lisa
    lisa2:47 PM

    @Capy captain=Mobile the TestFlight build is crashing on launch, take a look

  • MobileAPP2:47 PM

    Sent to Mobile.

    Open in Capy

From Slack

Add captain= to a message to Capy, and it goes to that Captain instead of starting a new thread.

9:41
A Captain in Capy at phone width, with the items waiting on you in its chat

From any screen

Open capy.ai on your phone to read the digest and answer. Add it to your Home Screen to get push notifications.

Briefcase

Multiplayer by default

Captains are shared with your team, and they talk to each other. Your Captain can hand work to a teammate’s Captain or ask it what only it knows.

A Captain’s Share dialog set to Only specific people, shared by its owner Maya Chen with lisa and Sam Patel

Shared Captains

Share a Captain with your whole organization or with specific people. Anyone it’s shared with can read it, message it and steer its threads.

The Platform Captain handing ACM-183 to the Mobile Captain, and the Mobile Captain replying that it took it

Captains talk to each other

Your Captain messages any Captain you can see to hand over work or ask a question, and the answer comes back to its thread.

Scientist capy

Orchestrations that run themselves

Ask a Captain to set up an automation, and it wakes on a schedule or when new work arrives, with nobody at the keyboard.

The Morning standup digest automation, active, on a weekday 9:00 AM schedule

On a schedule

“Every weekday at 9am, check the open PRs and update the digest.” The Captain sets it up and runs it.

New tickets reaching a Captain, and the Captain starting a thread for each

Work that arrives

An automation or a Slack message hands the Captain new tickets, and it starts a thread for each one.

Detective capy

Only what needs you

Each Captain keeps a digest of what’s waiting on you, what’s new and what’s in progress, and rewrites it as the work moves.

A Captain tagging its owner on the decisions that are theirs

Tags you when it’s yours

When a decision or a blocker is yours, the Captain tags you, and the tag notifies you like any mention.

A one-line reply to merge a pull request, and the Captain confirming it merged

Answer in one line

Reply “merge web#136” or “hold the migration until Monday”. The Captain acts on it and drops the item.

A Captain’s issues, each with an owner and a status

Tracked as issues

Work lands as issues with an owner and a status, so you can see where every piece stands.

Yuzu

Built to last

A Captain keeps its plan, rules and digest in its own files, so it continues where it stopped, days or weeks later.

A Captain’s files: its plan, rules, digest, skills and reports

Picks up where it stopped

The plan, the working notes and the digest are files the Captain rereads every turn, not a chat it has to remember.

The rules file every thread reads, open in the editor

Rules every thread reads

“From now on, run the e2e suite before calling a PR ready.” The Captain writes the rule, and every thread follows it.

Surprised capy

Captains beyond code

The same threads, schedules and reports run work that never opens a pull request: research, measurement, release notes and support.

lisa9:02 AM

Should we switch CI providers? Report back, no code.

Captain
  • Read14 months of CI runs
  • Compared3 providers on cost
  • Wroteci-providers.md

Research and reports

Send a question, not a ticket. Threads dig in, and the Captain hands back a written report with no code.

lisa9:02 AM

Every morning, score CI health and flag anything out of band.

Captain
CI red rate
3%
CI latency
11m
+18%
Oldest unmerged commit
4h
  • Created1 thread to fix CI latency

Daily scorecard

Every morning the Captain measures what you care about, flags anything out of band and starts a thread to fix it.

lisa9:02 AM

After each merge wave, post a changelog to #releases.

Captain
  • Read12 merged PRs
  • Draftedthe changelog
  • Postedto #releases

Changelogs in Slack

After each wave of merges, the Captain reads what shipped and posts the changelog to your channel.

lisa9:02 AM

Take every bug report from #support.

Captain
  • Reproduced5 reports
  • Filed3 issues, linked in #support
  • Asked2 reporters for steps

Support queue

Every bug report in your support channel gets reproduced, filed as an issue and linked back to the person who sent it.

FAQ

FAQ

Agent swarms, threads, machines, models and cost.

What is an agent swarm?+
An agent swarm is many AI agents working toward one goal at the same time, each on its own piece. In Capy, a Captain runs the swarm: it splits the goal into threads, each thread is a full coding agent with its own machine, and the Captain reads their reports and tells you what needs you.
When should I use a Captain instead of a thread?+
Start with a thread. Promote it to a Captain when you keep asking it to start other threads, or want one place to see the work across them. Use one Captain per area or feature.
How do I create a Captain?+
Choose New Captain and set its name, face, project, machine, model and access. You can also promote a thread you already have, or ask any thread: “Make a Captain called Mobile for the iOS and Android work.” In your first message, tell the Captain what it’s for.
Where do a Captain’s threads run?+
In the cloud by default, each thread on its own machine with your project’s repositories. The owner can send every thread to one paired computer instead, or name a machine for a single thread.
Which models can a Captain use?+
Threads use the Captain’s model unless you or the Captain pick another, so one Captain can mix models across its threads. They run on the subscriptions and API keys the Captain’s owner connected.
Can I steer a Captain from Slack or my phone?+
Yes. Add captain=<name> to a message to Capy in Slack and it goes to that Captain. On a phone, open capy.ai in the browser; on an iPhone or iPad, add it to your Home Screen from Safari to get push notifications.
Can I limit what a Captain does on its own?+
Yes. Give it a rule, such as “Ask before starting more than two threads at once” or “Nothing merges without my word.” It saves the rule in its own files and follows it from then on.
Who can see and message a Captain?+
Share a Captain with your whole organization or with specific people. Anyone it’s shared with can read it and its threads, message it and steer it, and other Captains that can see it can message it too. Its threads have the same access as the Captain, and the Captain always acts as its owner, whoever messages it.
What does a Captain cost?+
A Captain and its threads bill model usage like any thread, on your subscriptions and keys where you connected them. On a cloud machine they also bill machine time from your credits.

Hand off the whole roadmap

Promote any thread to a Captain, or ask Capy to make one. It starts the threads and tells you what needs you.