Salix

Agent runtimes

An agent runtime is the loop that reads a request, calls tools and writes the answer. Salix has its own, and it can also run external agent runtimes.

The built-in agent loop

By default, an agent uses Salix's built-in agent loop. It runs in the harness, outside the sandbox, and uses a cloud VM or a device only when the work needs a computer.

Codex and Claude Code

You can set an agent to use an external agent runtime, such as Codex or Claude Code. The agent then runs on the device you set, with that runtime.

Salix still delivers the agent's messages and keeps its conversation. The runtime does the work on the device.

Accounts and keys

An external runtime can use your own ChatGPT or Claude plan, or an API key. Salix keeps a reference to the account, not the credential itself.