Pre-Launch · Raising Pre-Seed

We don't write code.
We manage what does.

Orchestrate parallel AI coding agents in sandboxed environments. Each agent gets its own network namespace, VCS workspace, and database — fully isolated at the kernel level. No containers, no conflicts, no security risks.

The orchestration platform for agentic coding at scale.

Request Early Access See How It Works
∞ parallel agents
<1s environment startup
0 MB overhead per agent
1316+ tests passing
Launching 2026
Gleam + Rust + OTP + Nix
Built in Switzerland

Managing agents
is still chaos.

AI agents write 80% of new code. The $78B market built zero infrastructure for running them safely in parallel. No isolation, no coordination, no visibility. Every team hits the same walls.

Why agents need their own workspace
01

Environment Conflicts

Multiple agents sharing one filesystem. Merge conflicts, broken dependencies, wasted compute. Every. Single. Run.

02

Zero Observability

No real-time insight into what agents are doing. You wait for output, then debug after the fact — if you're lucky.

03

Uncontained Agents

Agents with full access to your network, credentials, and filesystem. No sandboxing, no boundaries. One rogue API call away from a security incident.

Your colonies,
orchestrated.

Kernel-level isolation, zero overhead, full observability — so you can run agents without worrying about security.

Learn more →

Parallel Agents

Run unlimited agents simultaneously. Each one sandboxed in its own network namespace — isolated at the kernel level.

agent-3
$ Refactoring auth...
agent-2
$ Running tests... ✓ 42 passed
agent-1
$ Building API... Adding route handler Generating types...
Learn more →

Live Preview

Watch agents work in real-time. Web, terminal, and logs — all in your browser.

Web Terminal Agent
Learn more →

Zero Containers

Linux network namespaces give you VM-level network isolation without Docker overhead. Instant startup, zero memory cost.

< 50ms cold start
ns-1
ns-2
ns-3
Learn more →

Jujutsu VCS

Every agent gets its own workspace. Automatic conflict resolution, no merge hell.

main agent-1 agent-2
Learn more →

Full Lifecycle

Spawn, monitor, review, and merge — from Stem or Bloom.

colony-alpha Running
colony-beta Building
colony-gamma Review

See It in Action

A unified control plane for every agent, colony, and brood — from high-level orchestration down to individual file diffs.

Alpha preview — UI is not final
Colony dashboard showing broods grouped by project with live status indicators
Dashboard

Broods grouped by project, live agent status, one-click creation.

Brood detail view showing parallel agent chat and colony preview cards
Brood Detail

Chat with your brood planner while agents work in parallel colonies.

Colony detail view showing API preview, agent stream, logs, and metadata
Colony Detail

Live service preview, agent conversation stream, logs, and metadata.

How It Works

1

Define

Create a colony.toml with your project config, agent instructions, and environment requirements.

2

Spawn

Colony sandboxes each agent in its own network namespace, database, and VCS workspace. Full kernel-level isolation — agents can't see each other or reach your host network.

3

Review & Merge

Monitor progress in real-time. Review changes, run tests, and merge — all from one dashboard.

Frequently Asked Questions

What AI agent does Colony use?
Colony uses Claude Code as its agent backend, managed via HumanLayer's hld daemon over JSON-RPC. Colony handles the orchestration, environment isolation, and lifecycle management — Claude Code handles the AI execution.
Why not just use Docker?
Docker adds ~200MB per container, takes seconds to start, and requires managing Dockerfiles. Colony uses Linux network namespaces — the same kernel primitive Docker uses internally — to provide equivalent network isolation without the overhead. Instant startup, zero memory cost, native filesystem performance.
Are AI agents secure in Colony?
Each agent runs in its own Linux network namespace — a kernel-level sandbox with its own network stack, IP address, and firewall rules. Agents can't see each other's traffic, can't reach your host network, and can't make unauthorized API calls. It's the same isolation you'd get from a VM, at zero overhead.
What is Jujutsu?
Jujutsu (jj) is a modern version control system that treats the working copy as a commit. This means every agent can work on its own branch without explicit branching, with automatic conflict detection and resolution.
Is Colony open source?
Colony will be released under the Functional Source License (FSL) — source-available from day one, with automatic conversion to Apache 2.0 after 2 years. The repository is currently private during pre-launch development. We'll open the source when we launch.
How does pricing work?
Colony uses a platform fee + usage model. Pay-as-you-go starts at $2.50/colony-hour with no platform fee. Pro ($149/mo) includes 100 hours at an effective rate of $1.49/hr. Volume discounts available for teams and enterprises.
What is AI agent orchestration?
AI agent orchestration is the coordination of multiple autonomous coding agents working in parallel. Colony provides the infrastructure layer — isolated environments, VCS workspaces, and real-time monitoring — so teams can run fleets of AI agents without conflicts.

Ready to orchestrate?

Colony is launching soon. Join the waitlist for early access and help shape multi-agent development.

No spam — we'll notify you when early access opens.

Read the documentation
Investor? We're raising a pre-seed round. See investor details