What CLI Workers Do

CLI workers turn Claude Code into a managed inference provider. Instead of each developer wiring their own local setup, you run dedicated CLI worker processes as shared infrastructure. Requests route through Auxot’s API, and the CLI worker handles the interaction with Claude Code.

CLI workers are second in routing priority after GPU workers, making them ideal for coding-focused tasks where tool use and agentic reasoning are important. Since Claude Code uses a fixed-cost subscription (see anthropic.com for current pricing), CLI workers provide predictable costs regardless of usage volume.

Setup

1. Create the CLI Provider and Worker Key

Create a CLI provider in Settings → Providers, then copy the worker key for that provider.

Or ask the admin agent:

Create a CLI provider for Claude Code and generate a worker key.

2. Run CLI Workers in Separate Containers or Hosts

CLI workers run separately from the Auxot server and connect back over WebSocket. You can run them on any machine that can reach your Auxot URL.

For install and runtime commands, see CLI Workers (OSS) →.

Headless Deployment

For servers and containers without a browser, generate a setup token on a machine with browser access:

claude setup-token

Then set the token on each CLI worker host/container:

export CLAUDE_CODE_OAUTH_TOKEN=<setup-token>

The CLI worker uses this token when it executes Claude Code requests.

Containerized Deployment

CLI workers are well-suited for containerized deployment, especially when you want elastic scaling.

Docker

Run CLI workers as separate containers and pass the worker key and router URL as environment variables.

ECS Fargate

For auto-scaling CLI workers on AWS:

{
  "containerDefinitions": [{
    "name": "auxot-cli-worker",
    "image": "your-registry/auxot-cli-worker:latest",
    "environment": [
      {"name": "AUXOT_ROUTER_URL", "value": "https://ai.yourcompany.com"},
      {"name": "AUXOT_GPU_KEY", "value": "from-secrets-manager"},
      {"name": "CLAUDE_CODE_OAUTH_TOKEN", "value": "from-secrets-manager"}
    ],
    "cpu": 1024,
    "memory": 2048
  }]
}

Each container registers as a separate worker with the router over WebSocket, and Auxot load-balances across connected workers.

Model Selection and Tool Control

CLI workers expose the models available through Claude Code. For available model IDs, the AUXOT_CLI_DEFAULT_MODEL environment variable, and tool control via AUXOT_CLI_TOOLS, see CLI Workers (OSS) →.

Specify a model in the API request, or let the router pick based on the worker’s default. For production deployments, grant the minimum set of tools needed for the worker’s intended use case.

Worker Isolation

Each CLI worker instance handles one request at a time by default. The router tracks each worker’s busy/idle state and only routes to idle workers.

For higher concurrency, deploy more CLI worker containers/hosts. Higher concurrency works well for lightweight tasks but may degrade quality on complex, multi-step workflows.