SELF-HOSTED · BRING-YOUR-OWN-KEYS · LOCAL-FIRST

Run a swarm of AI agents

on your machine.

SwarmLab is a self-hosted AI agent orchestration platform. Build agent teams, wire them into workflows, and ship real work — using your own provider keys, on your own hardware. Your code never leaves your box.

One-time $99 · Lifetime updates · Works with Claude, GPT & local models

localhost:3100/workflows/ship-feature
Ship Feature
workflow · 4 nodes · auto-saved
Nodes
Drag onto the canvas.
Transform
Agent Task
Invokes a configured agent…
If / Else
Routes the input to true or…
Switch
Routes input to one of up to 4…
Code
Run a JS expression or block a…
Trigger
Trigger
Manual trigger — kicks off…
Cron Trigger
Fires on a recurring schedule…
Agent Task
agent-task
Agent Task 2
agent-task
If
if-else
TRUEFALSE
HTTP Request
http-request
Approve
human-gate

WHAT IT IS

Three things, no surprises.

An orchestration platform.

Compose multi-agent workflows visually. Track work on a Kanban board.

Self-hosted by default.

Runs as a single binary on your machine. No SwarmLab cloud, ever.

Your keys, your bills.

Bring your own Anthropic, OpenAI, or local-model keys. Pay your provider directly.

USED BY DEVELOPERS WHO SHIP

3,418

stars on GitHub

1,250

in the Discord

782

self-hosted instances

localhost:3100/workflows/ship-feature
Ship Feature
workflow · 4 nodes · auto-saved
Nodes
Drag onto the canvas.
Transform
Agent Task
Invokes a configured agent…
If / Else
Routes the input to true or…
Switch
Routes input to one of up to 4…
Code
Run a JS expression or block a…
Trigger
Trigger
Manual trigger — kicks off…
Cron Trigger
Fires on a recurring schedule…
Agent Task
agent-task
Agent Task 2
agent-task
If
if-else
TRUEFALSE
HTTP Request
http-request
Approve
human-gate

WORKFLOWS

Drag nodes. Wire them up. Hit run.

Build repeatable agent pipelines on a visual canvas. Chain tasks together, add conditional branches, and drop in human approval gates — all without writing config files. Runs execute live and recover automatically if something crashes.

  • Drag-and-drop editor — no YAML, no config files
  • Mix agent work, API calls, code scripts, and human checkpoints
  • Run manually, on a schedule, or from a webhook

MISSION CONTROL

Watch every agent. Pause any of them.

One live dashboard for your entire swarm. See who's working, what they're saying, and how much they're spending — with streaming terminal output as agents work. If something goes sideways, Emergency Pause stops everything in one click.

  • Live output, status, and spend tracking for every agent
  • Active missions show progress — stuck ones float to the top
  • Emergency Pause stops all agents immediately
localhost:3100/mission-control
Mission Control
Live view of every active agent.
Emergency pause
Agents online
11
11 total
Active tasks
7
5 in-progress · 2 review
Active runs
3
currently running
Spend today
$4.82
across active runs
Recently active · 3last 60 min
PPepper
Live
Marketing & Branding
▸ LIVE OUTPUT
$ swarmlab task update SWM-013 \
  --status closed
Skills inventory delivered
  as a task comment for review.
Drafting copy for hero…
heartbeat 50s ago
PPeter
Live
Junior Developer
▸ LAST RUN · SWM-012
I'll research and document the
Model Control Protocol (MCP)
implementation for SwarmLab
agents. Let me start by gathering
info about the current agents
and their capabilities.
heartbeat 2h ago
BBruce
Idle
Code Reviewer
▸ LAST RUN · SWM-010
Steve's task (SWM-008) is
marked done. The re-review is
complete — all 9 items from my
original punch list were
correctly addressed in commit
'c81804a', with the 'NODE_ENV'
nit intentionally skipped.
heartbeat 5h ago
localhost:3100/agents/frontend-dev
Frontend Dev
agent · created 3 days ago
Role
Frontend
Model
claude-sonnet-4-6
Allowed paths
/src/**, /assets/**
Skills
code-reviewertest-runner+ add

AGENTS

Give each agent a role and boundaries.

Every agent gets its own model, personality, file permissions, and budget. Assign a code reviewer that only reads diffs, a designer scoped to /assets, or a researcher with read-only access. Attach skills to teach your team's conventions.

  • Choose from Claude, GPT, Codex, Ollama, or any OpenRouter model
  • Set allowed file paths and sandbox levels — enforced automatically
  • Daily budget caps so spending never surprises you

PRIVACY & CONTROL

Your code never leaves your machine.

Agents call your model providers directly using your API keys. There's no SwarmLab cloud in between — no telemetry, no data collection, no third party reading your code. If your laptop goes offline, SwarmLab keeps running for anything that doesn't need the model.

  • Zero data sent to SwarmLab — ever
  • Bring your own keys — pay providers directly at their prices
  • Each agent works on its own isolated git branch
no-egress.swarmlab.local
YOUR
Laptop
5 agents · 1 worktree
api.anthropic.comapi.openai.comlocalhost:11434
Anthropic
OpenAI
Ollama
No SwarmLab cloud in the middle. Direct provider calls only.

GETTING STARTED

From zero to a running agent in three steps.

1

Install

curl -fsSL get.swarmlab.dev | sh

One command. Single binary. No Docker, no Postgres.

2

Configure your provider

swarmlab config set anthropic.apiKey sk-ant-...

Bring your own API key. AES-256 encrypted on disk.

3

Start orchestrating

swarmlab start
# → Open localhost:3100

Web UI, CLI, and APIs — all running on your machine.

WHY SELF-HOSTED

What you get when the agents run on your machine.

CapabilityCloud agentsSwarmLab
Code stays on your machine
Bring your own provider keys
Pay providers at list price
Works offline
Multi-agent orchestration
Visual workflow buildervaries
Self-hosted on a workstation
Open audit of what's installed

Same orchestration UX. Different trust model.

One price. For life.

$99 one-time. Lifetime license, lifetime updates. No subscription, no seats, no per-token margin. You still pay your model providers directly at their list prices — SwarmLab takes nothing in the middle.

$99 USD · one-time