Build with AXL
AXL is a Rust platform for agents that do real work - answering from your documents, calling your systems, running multi-step processes - and that stay inside boundaries you define.
Start with what AXL is for the mental model and a map of everything you can attach to an agent. If you already know what you want, the tables below go straight there.
Start here
| I want to… | Read |
|---|---|
| Get AXL running and talk to an agent | Run AXL locally |
| Drive it from my terminal | Use the command line tool |
I want to build…
| I want to… | Read |
|---|---|
| Define an agent from configuration | Configure agents |
| Ground responses in my own documents | Use RAG |
| Compose reliable multi-step work | Build workflows |
| Give an agent tools or a sandbox | Connect tools |
| Connect Slack, Discord, Teams, or another channel | Add a channel |
| Publish an agent to another system | Use A2A or MCP |
| Add real-time speech and voice interaction | Add voice |
| Run coding work in an isolated factory | Use Workbench |
| Call AXL from an application |
I need my agent to…
| I need my agent to… | Read |
|---|---|
| Remember information across conversations | Long-term memory |
| Keep long conversations inside the model context window | Context and compaction |
| Delegate work to specialized agents | Subagents |
| Understand images supplied by users or tools | Multimodal agents |
| Use a specific model, provider, or reasoning level | Models and reasoning |
| Connect tools from an external MCP server | MCP tools |
| Run commands or manipulate files in an isolated workspace | Sandboxes |
| Present an interactive form or choice | Interactive MCP Apps |
| Run work on a recurring schedule | Schedules |
| Accept uploads and return durable artifacts | Files and artifacts |
| Require authentication and guard sensitive actions | Security |
| Track or cancel asynchronous work | Tasks |
| Trace runs, tool calls, latency, and cost | Observability |
| Diagnose a deployment or failed run | Troubleshooting |
Learn by building
Each of these takes a real job and builds it from an empty config directory, using whatever capabilities the job turns out to need.
| Walkthrough | What you build |
|---|---|
| Support agent | An agent that answers from your product docs, in Slack, and escalates when it should not guess |
| Document pipeline | A durable workflow that extracts structured data from uploads and routes exceptions to a person |
| Research agent | Parallel investigation that returns a sourced brief and files it as an artifact |
The examples are working configurations you can paste into your own
axl-config and adjust.
How this documentation is organized
Guides explain how to use a capability, with working examples, troubleshooting, and operational limits. Reference documents exact endpoints, configuration fields, and defaults. Internals describes implementation architecture for people changing AXL itself. The FAQ is indexed by the error message or symptom you are actually looking at.
Product requirements, comparison notes, reviews, and test plans stay in the repository; they are intentionally excluded from this site.