About

We built the chassis we kept having to rebuild.

Why this exists

In 2023, the team behind Agent Chassis was building production AI agents at three different companies. The pattern was identical every time: two engineers, three weeks, building the same scaffolding — retries, state management, tool registration, auth, logging.

We got to week four and realized we had a functioning chassis and no agent yet. The agent's actual job — the thing that justified the project — hadn't been written.

We shipped the chassis as an internal library. Six months later, engineers at other companies started asking for it. We open-sourced it. Agent Chassis was born from that repo.

The framework is MIT-licensed. Self-host for free, forever. The Cloud tier exists for teams who'd rather pay $79/mo than manage Redis and log aggregation themselves.

What we believe

AI agent development should feel like building a product, not maintaining infrastructure. The chassis layer — retries, state, auth, observability — is solved problem space. It should ship as a dependency, not as sprint work.

We believe in open-source infrastructure. The core framework will always be free and MIT-licensed. We make money on managed services for teams that want them, not by locking the primitives behind a paywall.

We believe production readiness is a first-class concern. A framework that makes demos easy but fails under load is a liability. Every default in Agent Chassis was set with production traffic in mind.

Get the framework

Principles

⚙️

Infrastructure, not product

The chassis exists so you don't have to think about it. Good infrastructure disappears.

🔓

Open by default

MIT-licensed core. No vendorlock. Run it anywhere, fork it freely, own your stack.

📐

Production-first defaults

Every config default is what you'd want at 10K requests/day. Not at localhost.

Build the agent, not the plumbing.

Free framework. Managed cloud when you're ready.

Get the framework