# Axon, a TypeScript framework for agent development

> Source: <https://axon.arclabs.it/docs/v2>
> Published: 2026-07-24 11:03:24+00:00

# What is Axon?

Axon is a runtime and toolkit for AI agents. It gives you a consistent way to define an agent, give it capabilities, keep its sessions, and run it locally or in the cloud.

You write the behaviour that is unique to your agent. Axon handles the runtime around it: context, tool calls, retries, sessions, and policy.

[A runtime, without the plumbing](#a-runtime-without-the-plumbing)

Define what an agent knows, what it can do, and what it is allowed to access. Axon then assembles the context, runs the loop, dispatches tools, persists sessions, and enforces the policy you set. The application boundary is a single call:

```
const { stream } = axon.stream({ prompt: [session, task], thread: "project-alpha" })
```

You receive a stream of entries and decide what your application does with them. The runtime concerns stay in one maintained system instead of being rebuilt for every agent.

[Add capabilities as modules](#add-capabilities-as-modules)

Integrations such as GitHub, Linear, Discord, and Stripe are modules. Install one and it brings its typed tools, client setup, event handling, and any required verification with it.

```
axon install @axon/github
axon install @axon/linear
```

Your agent can then use documented calls such as `github.openPr`

, `github.getPrDiff`

, and
`linear.createIssue`

. The integration is installed; it is not another subsystem for you to
maintain.

Browse the available modules in the [registry](/docs/agents).

[Work with the agent](#work-with-the-agent)

The terminal interface is where you run and inspect an agent. Tool calls are visible, sessions are navigable, and scripts are available without leaving the conversation.

Commands stay close to hand: `!`

runs a script, `@`

opens sessions, `*`

changes model, and
`>`

loads a prompt into the input.

It can also manage Ollama. Browse compatible local models, start a download, and switch between local and cloud inference without changing the agent.

[Deploy when it is ready](#deploy-when-it-is-ready)

```
axon deploy
```

Axon deploys the same project with a public URL, API key, and durable cloud storage. When
you need your own infrastructure, `axon build`

produces a Docker image for any container
platform.

An Axon agent is a project folder: its configuration, modules, and source travel together. It is not tied to a model vendor or a hosting provider.

Start with [Installation](/docs/v2/getting-started/installation), then build [your first
agent](/docs/v2/getting-started/first-agent).