Free Code Claude CLI

free-code Claude CLI

Skill by ara.so — Daily 2026 Skills collection.

free-code is a buildable fork of Anthropic's Claude Code CLI — a terminal-native AI coding agent. It removes all telemetry and phone-home callbacks, strips Anthropic's injected system-prompt guardrails, and unlocks 54 experimental feature flags that are disabled in the public npm release. It supports five model providers: Anthropic, OpenAI Codex, AWS Bedrock, Google Vertex AI, and Anthropic Foundry.

Installation

One-liner (recommended)

curl -fsSL https://raw.githubusercontent.com/paoloanzn/free-code/main/install.sh | bash

This installs Bun if needed, clones the repo, builds with experimental features, and symlinks free-code on your PATH.

Manual build

# Requires Bun >= 1.3.11
curl -fsSL https://bun.sh/install | bash

git clone https://github.com/paoloanzn/free-code.git
cd free-code
bun run build
./cli

Key CLI Commands

# Start interactive REPL
free-code

# One-shot prompt
free-code -p "explain this codebase"

# Specify model
free-code --model claude-opus-4-6

# OAuth login (runs /login slash command)
free-code /login

# Run from source (slower, no build step)
bun run dev

Slash commands inside the REPL

/login          Authenticate with your chosen provider
/help           List all available commands
/clear          Clear conversation context
/compact        Compact context window
/memory         View/edit memory files

Build Variants

Enable specific feature flags

# Unlock ultrathink + ultraplan only
bun run ./scripts/build.ts --feature=ULTRATHINK --feature=ULTRAPLAN

# Dev build with bridge mode added
bun run ./scripts/build.ts --dev --feature=BRIDGE_MODE

# Full experimental unlock
bun run build:dev:full

Model Provider Configuration

Switch providers entirely via environment variables — no code changes needed.

Anthropic (default)

export ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY"
free-code --model claude-opus-4-6

Available models: claude-opus-4-6, claude-sonnet-4-6, claude-haiku-4-5

Override model defaults:

export ANTHROPIC_DEFAULT_OPUS_MODEL="claude-opus-4-6"
export ANTHROPIC_DEFAULT_SONNET_MODEL="claude-sonnet-4-6"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="claude-haiku-4-5"
export ANTHROPIC_BASE_URL="https://api.anthropic.com"  # custom endpoint

OpenAI Codex

export CLAUDE_CODE_USE_OPENAI=1
free-code --model gpt-5.3-codex

Available models: gpt-5.3-codex, gpt-5.4, gpt-5.4-mini

AWS Bedrock

export CLAUDE_CODE_USE_BEDROCK=1
export AWS_REGION="us-east-1"
# Uses standard AWS credential chain: env vars, ~/.aws/config, or IAM role
free-code

Additional Bedrock variables:

export ANTHROPIC_BEDROCK_BASE_URL="https://..."   # custom endpoint
export AWS_BEARER_TOKEN_BEDROCK="$TOKEN"           # bearer token auth
export CLAUDE_CODE_SKIP_BEDROCK_AUTH=1             # skip auth (testing only)

Google Cloud Vertex AI

# Authenticate first
gcloud auth application-default login

export CLAUDE_CODE_USE_VERTEX=1
free-code

Anthropic Foundry

export CLAUDE_CODE_USE_FOUNDRY=1
export ANTHROPIC_FOUNDRY_API_KEY="$ANTHROPIC_FOUNDRY_API_KEY"
free-code --model my-deployment-id

Experimental Features Reference

Build with bun run build:dev:full to unlock all 54 flags, or pass --feature=FLAG individually.

High-value flags

Environment Variables Quick Reference

# Authentication
ANTHROPIC_API_KEY              # Anthropic API key
ANTHROPIC_AUTH_TOKEN           # Alternative auth token
CLAUDE_CODE_OAUTH_TOKEN        # OAuth token via env (skips /login)

# Provider selection
CLAUDE_CODE_USE_OPENAI=1       # Switch to OpenAI Codex
CLAUDE_CODE_USE_BEDROCK=1      # Switch to AWS Bedrock
CLAUDE_CODE_USE_VERTEX=1       # Switch to Google Vertex AI
CLAUDE_CODE_USE_FOUNDRY=1      # Switch to Anthropic Foundry

# Model overrides
ANTHROPIC_MODEL                # Override default model for session
ANTHROPIC_BASE_URL             # Custom API base URL

# Tuning
CLAUDE_CODE_API_KEY_HELPER_TTL_MS  # API key helper cache TTL in ms

Project Structure

scripts/
  build.ts              # Build script — feature flag injection lives here

src/
  entrypoints/cli.tsx   # CLI entry — Commander.js setup, arg parsing
  commands.ts           # Slash command registry
  tools.ts              # Agent tool registry
  QueryEngine.ts        # Core LLM query orchestration
  screens/REPL.tsx      # Interactive terminal UI (React + Ink)

  commands/             # /slash command implementations
  tools/                # Agent tools: Bash, Read, Edit, Glob, Grep, etc.
  components/           # Ink/React terminal UI components
  services/
    api/                # Anthropic + Codex API clients
    oauth/              # OAuth flows (Anthropic, OpenAI)
  state/                # Global app state store
  utils/model/          # Model config, provider routing, validation
  bridge/               # IDE bridge (BRIDGE_MODE)
  voice/                # Voice input (VOICE_MODE)
  skills/               # Skill system
  plugins/              # Plugin system
  tasks/                # Background task management

Common Patterns

Use ultrathink for hard problems

After building with --feature=ULTRATHINK:

> ultrathink refactor this authentication module to use JWTs

The keyword ultrathink in your prompt triggers extended reasoning mode.

Pipe files into one-shot mode

cat src/auth.ts | free-code -p "find security issues in this file"
free-code -p "write tests for $(cat src/utils.ts)"

Run with a specific provider per session

# One-off Bedrock session without changing env permanently
CLAUDE_CODE_USE_BEDROCK=1 AWS_REGION=eu-west-1 free-code

# One-off Vertex session
CLAUDE_CODE_USE_VERTEX=1 free-code -p "review this PR diff"

Custom build for CI with minimal features

// scripts/build.ts accepts --feature flags
// Build only what you need for a headless CI agent:
// bun run ./scripts/build.ts --feature=VERIFICATION_AGENT --feature=AGENT_TRIGGERS

Troubleshooting

bun: command not found

curl -fsSL https://bun.sh/install | bash
source ~/.bashrc  # or ~/.zshrc

free-code: command not found after install

The install script symlinks to your PATH. If it's missing:

cd free-code
ln -sf "$(pwd)/cli" /usr/local/bin/free-code

Build fails on experimental flags

Some of the 88 flags don't compile cleanly. Use build:dev:full which only enables the 54 working flags. For broken flags, check FEATURES.md for reconstruction notes: