"It's turtles all the way down."
Turtlez is an Node-based implementation of the Recursive Language Model (RLM), originally developed in Python. It provides a robust, agentic coding experience designed to handle infinite context by recursively managing memory and tool-use.
The goal of Turtlez is to transition from a simple RLM chat interface into a fully Agentic Coding Environment. By leveraging recursive context management, Turtlez allows AI agents to maintain coherence over massive codebases, providing a truly "infinite" context window.
- Paper: Recursive Language Models
- Original Implementation: RLM REPL)
- Phase 1: Core Backend (Context Store, SQLite Integration)
- Phase 2: RLM Engine & Recursive Orchestration
- Phase 3: Real-time SSE API Routes
- Phase 4: Next.js Chat Interface & Trace Inspector
- Phase 5: Agentic Coding Tools (File System Access, Shell REPL)
- Phase 6: Autonomous Workspace Management
- Phase 7: Collaborative Multi-Agent Loops
- Node.js (v20+)
- Bun (recommended) or NPM
cd rlm-chat
bun installcd rlm-chat
# Rebuild sqlite for your platform if needed
npm rebuild better-sqlite3
# Start dev server
bun run dev- Framework: Next.js 15 (App Router)
- Runtime: Node.js / Bun
- Database: SQLite (better-sqlite3)
- Styling: Tailwind CSS v4
- Testing: Playwright (E2E), Vitest (Unit)
We love contributions! Please see our CONTRIBUTING.md for guidelines on our "Blog Notebook" strategy for commits and PRs.
Distributed under the MIT License. See LICENSE for more information.
The name is a reference to "Turtles all the way down," an expression of the infinite regress problem. In our context, it represents the recursive nature of the language model—using its own previous outputs and external context to build an infinite tower of reasoning. It's a joke, an inside one, and a philosophy all in one.