01
You're the switchboard
A window open for every session, and you're the wire between them — copy from this one, paste into that one, remember which is waiting on you. The coordination is a job, and it's yours.
Independence Day 2026 · open source · v0
flotilla is a drop-in chief of staff
for the AI coding agents you already run.
Stop shuffling 10 terminal windows. flotilla turns separate Claude Code / Codex / Grok sessions into one centrally coordinated fleet. A single hub agent coordinates multiple streams of work and lets you drive strategy, not implementation.
flotilla your fleet
Chief of staff
Your chief of staff
on itclaude-code
Flotilla · led by an XO
Product flotilla
on itproduct-xo
Flotilla · led by an XO
Research flotilla
needs youresearch-xo
Desk · backend
Backend
idlegrok
Desk · frontend
Frontend
workingcodex
Desk · data
Data
workingcodex
Desk · infra
Infra
idlegrok
works with
Claude Code
Codex
Grok
Cursor · on the roadmap
01 — the problem you already have
One coding agent was a superpower. Five is a second job.
You pay for Claude Code. Maybe Codex and Grok too. Each one, on its own, is brilliant. Running several at once is where it falls apart — and if that's your day, you already know exactly how.
02
They forget everything
Every new session starts from zero. The conventions you set, the decisions you made yesterday, what worked last time — gone. You re-explain your own project, again, to the same tool.
03
Nothing moves unless you watch
Look away and progress stops. So you babysit — not because the work is hard, but because there's no one minding the fleet while you step out for lunch.
02 — a day with flotilla
Picture the same day, run from one chat channel.
Nothing here needs a terminal open. You talk to one agent — your chief of staff — and it runs the rest.
8:12 · coffee
You skim one brief, not ten sessions
You open the chat channel on your phone. Your chief of staff has a two-line summary waiting: three things shipped overnight, one needs your call. You reply “go with option B” and put the phone down. No terminal was opened.
11:40 · heads-down
Your desks work while you do your own thing
You're deep in your own editor. Meanwhile your backend desk is running the migration you queued and your docs desk is rewriting a guide. You're not watching either — and they don't need you to be.
1:15 · lunch
A result finds you, and you answer in a line
Your research desk finishes a benchmark; the coordinator posts the result to your phone. You read it in the lunch line and type “ship it.” It routes to the right desk and you get a confirmation that the message actually landed — not a hope that it did.
4:30 · new work lands
A fresh desk that already knows your project
A new project comes in. You spin up a desk for it in minutes over a coding agent you already run — and it starts already knowing your conventions and recent decisions, from a memory that carried over, instead of from a blank slate.
6:00 · signing off
You close the laptop; the fleet keeps its own clock
You log off. The coordinator keeps advancing the work you already approved and stops to ask only when something is genuinely yours to decide. Tomorrow's brief is already being written.
03 — what you get
Three ideas do the whole job. Here they are in plain words.
01
Flotillas — your domains, run for you
Group your work into domains — backend, research, docs, infra — and each gets its own small crew of agents that runs mostly on its own. Start one when a project lands, retire it when it's done, so the shape of your fleet follows the work instead of a flat pile of to-dos. (A flotilla is one such coordinated domain — the product is named for them.)
02
Memex — it remembers, and it learns
Each desk keeps its own memory — your conventions, the decisions you've made, what worked last time — so it doesn't reset to zero every session. That memory is memex, a companion project flotilla pairs with. Learning across sessions is the point.
03
A chief of staff — one point of contact
You don't manage five agents. You talk to one — the coordinator. It takes your intent, routes it to the right desk, chases the replies, and reports back in plain language. One conversation; the whole fleet behind it.
04 — the part that makes it easy
You keep the tools you already run. Nothing to rebuild.
flotilla doesn't replace Claude Code, Codex, or Grok — it sits over them. There's no new agent framework to learn and no migrating your work into someone else's runtime. You point the coding agent you already run at one file, and you're coordinated in minutes.
Starting over on a framework
- Rebuild your workflow inside a new agent runtime
- Re-teach it the tools you already trust
- Learn a new way of working before anything ships
Dropping flotilla in
- Keep the exact coding agents you run today
- Point one of them at a setup file — it installs the rest
- Drive the fleet from chat within minutes
Curious how flotilla compares to agent frameworks and terminal tools? agent platforms vs flotilla · herdr vs flotilla
05 — how it actually works
One hub you talk to. Your desks around it. Everything on the record.
-
you talk to one agent
Chat is the whole interface
You message the coordinator like a colleague — “what shipped overnight?”, “benchmark the cache and report back.” From your laptop or your phone. That's the entire cockpit.
-
it reaches your real sessions
Delivered, and confirmed
The coordinator hands work to a desk running in a real session, and tells you the message actually landed. If a desk has crashed, it says so — it never quietly drops what you asked for.
-
nothing gets lost
Every message on the record
Each instruction and reply is copied to your chat channel under the sender's name — a timestamped history you can read back from any device, so the log never disagrees with what happened.
-
it runs safely on its own
Autonomy with a leash
Each desk acts unattended on safe work and stops for your OK on the risky kind. Progress without babysitting; a hand on the brake when it matters.
✓ real today
- Mixed agents, one fleet — Claude Code, Codex, and Grok desks together
- One coordinator — routes work, confirms delivery, reports back
- The full record — every message mirrored to your chat channel
- A fleet that keeps its own clock — advances approved work, flags a stuck desk
- Drive it from your phone — message the fleet from the channel
- A local fleet board —
flotilla dash, your desks and work at a glance
◌ in trial · on the roadmap
- A Codex-run coordinator — code-complete, in supervised trial
- The goals map — the domain view pictured up top, as a first-class dashboard
- A Cursor desk — the next coding agent we're adding
- Richer reporting — more than the chat mirror
v0, built in the open — so you can see exactly what's real and what's next before you rely on it.
06 — the one thing you do next
Don't set it up yourself. Tell your agent to.
You already run a coding agent that can install software and follow a guide. So hand it the guide. Paste this into Claude Code, Codex, or Grok — it does the rest and explains each step as it goes.
Paste into your coding agent:
Read https://github.com/jim80net/flotilla/blob/main/llm.md and set flotilla up for me: install it, tag one of the coding agents I already have running, deliver a first message to it, and start the coordinator. Explain each step in plain language as you go, and stop and tell me if anything is missing.
That guide walks the whole thing end-to-end — checking prerequisites, installing, tagging your first agent, sending your first message, and starting the self-continuing clock. Wiring the chat channel is an optional last step.