Coderix

A fully open-source (Apache 2.0) terminal AI programming assistant — the free alternative to Claude Code.

Coderix is a powerful AI coding agent that runs in your terminal or as a desktop app. It can read, write, edit files, execute shell commands, search code, and more — all through natural language conversation. Built with Ink/React for a beautiful TUI experience, plus an Electron desktop app (React DOM + Monaco + xterm) that shares the same core engine.

---

Why Coderix?

	Claude Code	Coderix
License	Proprietary	Apache 2.0
Source	Closed	Fully open
Provider	Anthropic only	Anthropic / DeepSeek / OpenAI
Pricing	Per-token billing	Bring your own key
Extensibility	Limited	Full plugin architecture

---

Quick Start

Prerequisites

• Node.js >= 22
• An API key from DeepSeek, Anthropic, or OpenAI

Install

git clone https://github.com/AgenticMatrix/coderix.git
cd coderix
./install.sh --local

Development

Coderix ships two interfaces backed by the same core engine:

Command	Interface
npm run dev:cli	TUI — the terminal interface (Ink/React)
npm run dev:desk	Desktop — the Electron app (React DOM)

# TUI (terminal) version
npm run dev:cli

# Desktop (Electron) version
npm run dev:desk

# Desktop (Electron) version — one-click launcher, auto-frees port 5173
./start_desk.sh

Configure

# First-time setup wizard
coderix setup

# Or manually edit ~/.coderix/settings.json

Start Coding

# Interactive session
coderix

# One-shot query
coderix --print "Explain the src/core/query-engine.ts file"

# Switch model
coderix --model
coderix -m "deepseek/deepseek-v4-pro"

---

Features

• Beautiful TUI — Built with Ink + React 19, full terminal rendering
• Multi-Provider — Anthropic (Claude), DeepSeek, OpenAI-compatible endpoints
• 15+ Tools — read, write, edit, bash, grep, glob, web-fetch, web-search, task management, todo
• Streaming Tool Queue — Tools enqueue and execute as they are parsed from the LLM stream, with bounded concurrency (default 32)
• Streaming — Real-time text, thinking, and tool-use streaming via ContentBlock events
• Agent Loop — Autonomous multi-turn reasoning with tool call execution
• Permission System — plan / ask / auto modes with risk-level classification
• Context Management — Token budget tracking and automatic compaction
• Hook System — Extensible lifecycle hooks
• Skills — Pluggable skill modules
• Session Management — Checkpoint, resume, fork sessions
• Model Picker — Interactive terminal model selection (coderix --model / coderix setup)
• Coordinator Mode — Orchestrate a team of worker sub-agents instead of editing files directly
• Desktop App — Electron desktop client with Monaco editor, xterm terminal, and source control

---

Keyboard Shortcuts

Shortcut	Action
Enter	Send message
Escape	Clear input / close sub-agent view
Ctrl+C	Interrupt agent / kill sub-agents / clear input / double-press to exit
Ctrl+B	Move sub-agent to background (unblocks main agent, agent keeps running)
Ctrl+T	Toggle sub-agent transcript view
Ctrl+P	Toggle task & todo panels
Ctrl+O	Toggle expand/collapse all blocks
Ctrl+K	Toggle team picker
Ctrl+Enter	Insert newline
↑ / ↓	Navigate input history
← / →	Move cursor
Tab	Auto-complete slash command
PageUp / PageDown	Freeze / unfreeze display

---

Configuration

Edit ~/.coderix/settings.json:

{
  "model_list": [
    {
      "model": [
        {
          "name": "deepseek-v4-pro",
          "price": {
            "input": 3,
            "cache_read_input": 0.025,
            "output": 6,
            "currency": "CNY",
            "unit": 1000000,
            "concurrency": 500,
            "max_context": 1000000
          }
        },
        {
          "name": "deepseek-v4-flash",
          "price": {
            "input": 1,
            "cache_read_input": 0.02,
            "output": 2,
            "currency": "CNY",
            "unit": 1000000,
            "concurrency": 2500,
            "max_context": 1000000
          }
        }
      ],
      "provider": "deepseek",
      "base_url": "https://api.deepseek.com/anthropic",
      "auth_token_env": "YOUR_DEEPSEEK_API_KEY"
    },
    {
      "model": [
        "claude-sonnet-2025",
        "opus-4.8"
      ],
      "provider": "anthropic",
      "base_url": "https://api.deepseek.com/anthropic",
      "auth_token_env": "YOUR_ANTHROPIC_API_KEY",
      "price": {
        "input": 3,
        "output": 15,
        "currency": "USD",
        "unit": "1M tokens"
      }
    },
    {
      "model": [
        "gpt-5",
        "gpt-5-mini"
      ],
      "provider": "openai",
      "base_url": "https://api.openai.com/v1",
      "auth_token_env": "YOUR_OPENAI_API_KEY"
    }
  ],
  "default_model": "deepseek/deepseek-v4-pro",
  "max_tool_concurrency": 32,
  "theme": "dark"
}

---

Coordinator Mode

Coordinator mode transforms Coderix from a direct code-editing agent into a pure orchestrator. Instead of reading files and writing code itself, the coordinator decomposes tasks, spawns parallel worker sub-agents to do the work, and synthesizes results.

Enable

# Via environment variable
export CODERIX_COORDINATOR=true
coderix

# Or persist in settings
coderix config set coordinator_mode true

How It Works

The coordinator follows a 4-stage workflow:

Phase	Who	What
Research	Workers (parallel)	Explore the codebase, find relevant files, understand the problem
Synthesis	Coordinator	Read findings, design the approach, write implementation specs
Implementation	Workers	Make targeted changes per spec
Verification	Workers (fresh)	Independently test and review changes

The coordinator only has orchestration tools (Agent, SendMessage, TaskGet, TaskStop, TeamCreate, TeamDelete, Sleep) — no filesystem or code-editing tools. All file operations are delegated to workers.

---

CLI Reference

Command	Description
coderix	Start interactive session
coderix "query"	One-shot question
coderix --help	Show help
coderix --version	Print version
coderix --model	Interactive model picker
coderix -m "provider/model"	Set model directly
coderix setup	First-time setup wizard

---

License

Apache 2.0 — fully open source. Use it, modify it, ship it.