@flue/cli

CLI for running Flue AI-enabled workflows, locally or in GitHub Actions.

Install

bun install @flue/cli
npm install @flue/cli
pnpm install @flue/cli

Usage

flue run <workflowPath> [--args <json>] [--model <provider/model>] [--sandbox <image>]

flue run .flue/workflows/triage.ts
flue run .flue/workflows/triage.ts --model anthropic/claude-sonnet-4-5
flue run .flue/workflows/triage.ts --args '{"issueNumber": 123}'
flue run .flue/workflows/triage.ts --sandbox my-org/my-sandbox:latest

The CLI auto-starts an OpenCode server if one isn't already running. The opencode binary must be installed and on PATH.

Sandbox Mode

The --sandbox <image> flag runs the OpenCode server inside a Docker container for security isolation. The LLM and its tool calls execute inside the container, while the host retains control of secrets (like API keys). Shell commands via flue.shell() also execute inside the container.

Prerequisites: Docker (GitHub Actions supported). Your container image must have OpenCode and git installed, and should start the OpenCode server on port 48765. Any other tools your workflows need (e.g., curl, pnpm) can be added to the image as well.

Model Configuration

The CLI uses the local OpenCode server's model configuration. Either:

• Pass --model to the CLI: flue run workflow.ts --model anthropic/claude-sonnet-4-5
• Or set "model" in your project's opencode.json

Provider API keys (e.g. ANTHROPIC_API_KEY, OPENAI_API_KEY) are read from the environment at runtime.