opencode-rlm

AI-powered development tool with TUI interface and Recursive Language Model (RLM) context management for unlimited conversation context.

Features

• RLM Context Management: Automatic context archival and retrieval enables conversations beyond context window limits
• Multi-Provider Support: Anthropic (Claude), OpenAI (GPT-4), and more
• Rich TUI Interface: Full-featured terminal UI with Textual
• 14 Built-in Tools: bash, read, write, edit, glob, grep, todo, webfetch, memory, and more
• Session Management: Fork, export, and manage conversation sessions
• Slash Commands: /context, /search, /compact, /models, /agents

Installation

via npm (recommended)

npm install -g opencode-rlm

via pip

pip install opencode-rlm

From source

git clone https://github.com/tekcin/opencode-rlm.git
cd opencode-rlm
pip install -e .

Requirements

• Python 3.11+
• Node.js 16+ (for npm installation)

Usage

# Start the TUI
opencode

# Start with a specific project
opencode /path/to/project

# Continue last session
opencode --continue

# Run a single command (non-interactive)
opencode run "explain this codebase"

# List available models
opencode models

Configuration

Create ~/.config/opencode/config.json:

{
  "model": "anthropic/claude-sonnet-4",
  "theme": "opencode",
  "rlm": {
    "enabled": true,
    "threshold_ratio": 0.33,
    "auto_retrieve": true
  }
}

Environment Variables

export ANTHROPIC_API_KEY=sk-ant-...
export OPENAI_API_KEY=sk-...

RLM (Recursive Language Model)

RLM enables unlimited context by automatically:

1. Tracking token usage across the conversation
2. Archiving older context when threshold is reached (default: 33%)
3. Retrieving relevant archived context for new queries
4. Injecting context into the conversation seamlessly

Manual RLM Commands

• /context - Show current token usage
• /search <query> - Search archived context
• /compact - Manually trigger context archival

RLM Tools

• memory - Search and retrieve archived context
• rlm_query - Execute Python queries on archived context

License

MIT

Author

Michael Thornton (tekcin@yahoo.com)