Infrastructure for AI coding agents

Describe a task, walk away.
Come back to a PR that followed your rules.

Lotsa pilots your coding agents through a rule-enforced pipeline, running in parallel across your repos. Self-hosted. Your keys, your rules.

Get started on GitHub
$ pip install lotsa
You're on the list — we'll email you when Teams ships.
or get notified about the upcoming Lotsa Teams launch →

Apache-2.0 · self-hosted · works with Claude Code & Codex · your keys never leave your machine

lotsa · fleet piloting
Fleet · 5 tasks
{{ tk.title }}
{{ tk.project }} {{ tk.subStep }}
{{ tk.stepText }} {{ tk.pillText }}
Activity Artifacts Changes
+ if (!token) return res.status(401)
− // TODO: handle auth later
{{ selTitle }} {{ selProject }} {{ prBadge }}
{{ st.mark }}{{ st.label }}
{{ watchPill }} watching {{ prBadge }}
{{ c.tag }} {{ c.text }}
a fleet of tasks, each piloted to merge — watched all the way into harbour
The shift tasks, not chats servers, not terminals async, not synchronous a fleet, not a single agent a pipeline, not freeform
Determinism

Your rules, enforced by the system — not the model's memory.

Conventions you set become stages the pipeline must execute. They don't depend on a prompt the agent might forget three tasks from now.

Rules live in the pipeline

Your guardrails and review steps are pipeline stages, executed in order — every run, not when the model remembers.

Gates, not vibes

Each stage must pass before the next begins. No skipped tests, no silent edits, no "I forgot to write the spec."

Humans hold the keys

Approve the plan. Approve the PR. Lotsa never merges on its own — you stay the pilot at every gate you choose.

Task #100 gets the same rigor as task #1.
Capabilities

What it does

01

Orchestrate agents

Drive Claude Code, Codex, or your own model through one consistent, repeatable flow.

02

Run in parallel

Kick off many tasks at once across every repo you own — a whole fleet moving together.

03

Enforce a pipeline

Spec, plan, test, code, review, verify, PR — in that order, on every single run.

04

Human gates

Pause for approval exactly where you want a person in the loop. Skip them where you don't.

05

Bring your own keys

Self-hosted and model-agnostic. Nothing routes through us — ever.

06

End in a PR

Every task lands as a branch and a pull request — then follows it to merge, fixing review and CI feedback along the way.

The run

A task, start to finish

1

Describe it

Tell Lotsa the task in plain language, right in the chat.

2

Spec & plan

It writes a spec and a plan against your rules — and waits for your gate if you want one.

3

Build & test

Your agent writes the code and the tests, iterating until everything is green.

4

Review & verify

A self-review pass and your human gate stand between the work and the branch.

5

Open the PR

Lotsa pushes a branch and opens a pull request — your harbour for the finished task.

Then multiply it

Now run a whole fleet at once.

The same pipeline, many tasks, every repo — each one piloted into its own pull request in parallel.

api rate-limit the gateway
web dark-mode tokens
auth migrate to OAuth 2.0
Beyond the PR

It doesn't stop at the PR.

Most tools hand you a pull request and walk away. Lotsa keeps watching it — when a reviewer or CI leaves actionable feedback, it dispatches a fix, pushes again, and re-runs its own review, automatically, until the PR is merged. You're not the one ferrying review comments back to an agent.

It pilots the ship all the way into harbour — not just to the harbour mouth.

PR #128 opened watching review / CI feedback dispatch fix → push
re-runs its own review and repeats — automatically — until every check passes
Merged into main
When to reach for it

Where it fits — use it alongside your other coding tools.

Lotsa doesn't replace the tools you code with. Reach for it when you want…

enforced rules — your conventions run as steps, every task, not when the model remembers

quality you can trust — tests written, the diff reviewed by a fresh agent, the build verified before any PR opens

to close your laptop — kick it off, walk away, come back to an open pull request

to run a fleet — many tasks at once, across every repo you own

the PR handled to merge — it watches the PR and fixes review/CI feedback for you

a person in the loop only where it counts — a gate where you want one, nowhere you don't

on your own terms — your machine, your keys; nothing routes through anyone

"I keep Claude Code open for hands-on work, and hand Lotsa the tasks I want run to a standard without babysitting. Use the right one for the moment."
— from the maintainer
stage 01

You describe

An idea, a ticket, a chore.

stage 02

Your agents work

Claude Code, Codex, your model.

lotsa pilots

The pipeline

Rules, gates, order — enforced.

stage 04

Harbour

A pull request, ready to review.

Self-hosted

Runs on your machine. Answers to no one but you.

One binary, static-hostable, Apache-2.0. Read every line, run it air-gapped, bring your own model keys.

Your API keys stay local
Your code never leaves your infra
Apache-2.0 — audit the whole thing
Coming soon

Lotsa Teams is coming.

Shared pipelines, org-wide rules, and a fleet view for everyone. Be first to know when it ships.

No spam. One email when Teams launches.