Overview

Agents are the coding assistants CodeFreedom orchestrates. Each agent runs either natively on your machine or inside a sandboxed Docker container, with automatic proxy configuration and model routing.

Available Agents¶

Agent	CLI Name	Alias	Description
Claude Code	`claude-code`	`cc`	Anthropic's coding agent — native Claude models or proxy-routed
MiMo Code	`mimo-code`	`mc`	Xiaomi's coding agent — 0-click proxy config with model auto-discovery
OpenCode	`open-code`	`oc`	Terminal-native AI coding agent — 0-click proxy config with model auto-discovery
Pi Code	`pi-code`	`pc`	Earendil's AI coding agent — extension-based dynamic model discovery

Quick Reference¶

cf run agent claude-code          # Launch Claude Code
cf run agent mimo-code            # Launch MiMo Code
cf run agent open-code            # Launch OpenCode
cf run agent pi-code              # Launch Pi Code
cf run agent list                 # List all available agents

Short aliases:

cf r ag cc                       # Claude Code
cf r ag mc                       # MiMo Code
cf r ag oc                       # OpenCode
cf r ag pc                       # Pi Code
cf r ag list                     # List agents

How Agents Work¶

When you launch an agent, CodeFreedom:

Loads your profile — reads environment variables from ~/.codefreedom/profiles/<name>.yaml
Generates agent config — auto-detects the proxy and builds the agent's config file with all available models
Starts MCP tools — ensures requested tool containers (Chrome, Web, GitHub) are running
Launches the agent — either natively or in a sandboxed Docker container

Native vs Sandbox Mode¶

Mode	Flag	Description
Native	(default)	Runs directly on your host machine
Sandbox	`--sandbox`	Runs inside an isolated Docker container

Sandbox mode is recommended for untrusted code. Native mode is simpler and has no container overhead.

GPU Support¶

Sandbox mode supports GPU acceleration:

cf r ag cc --sandbox --cuda      # NVIDIA GPU (CUDA)
cf r ag cc --sandbox --rocm      # AMD GPU (ROCm)

Common Flags¶

All agents share these flags:

Flag	Description
`-p`, `--profile NAME`	Load a named profile (default: `default`)
`-l`, `--list-profiles`	List available profiles and exit
`--sandbox`	Run inside a Docker container
`--run-as-me`	Run sandbox as host user (uid/gid match)
`-- <args>`	Forward arguments to the agent

Profiles¶

Each agent reads its configuration from a profile. Profiles define which model, provider, and environment variables the agent uses.

cf r ag cc -p deepseek           # Use the 'deepseek' profile
cf r ag mc -p costeffective      # Use the 'costeffective' profile

See Recipes for pre-configured profiles.

Tool Integration¶

Agents connect to MCP tools automatically when launched. Tools are persistent Docker containers — start once, use from any agent session.

cf run tools start                # Start all tools
cf r ag cc                       # Launch agent — tools are auto-connected

See Tools for available tools and configuration.