The Problem
Developers working with multiple AI coding assistants often juggle provider-specific CLIs, API keys, and authentication flows. claw-dev addresses this by providing a single launcher for a bundled terminal client, letting users route model requests to Anthropic, OpenAI, Gemini, Groq, OpenRouter, Ollama, Copilot, or z.ai without switching tools.
What This Does
claw-dev is a local multi-provider coding assistant launcher. The core is a bundled terminal client in Leonxlnx-claude-code/, which is a modified version of Claude Code (the repo includes anthropic-ai-claude-code-2.1.87.tgz and a patch-branding.js script). The launcher scripts (claw-dev-launcher.js, claw-dev.sh, claw-dev.cmd, claw-dev.ps1) provide platform-specific entry points.
The architecture centers on an Anthropic-compatible proxy (src/anthropicCompatProxy.ts, referenced in the README but not present in the file listing) that translates requests from the Claude Code client to non-Anthropic providers. The src/bridge/ directory contains the client-server communication layer, while src/commands/ holds a large set of CLI subcommands (e.g., agents, branch, config, doctor, mcp). The src/buddy/ directory includes a companion sprite UI component, suggesting a visual assistant feature.
The project is TypeScript-heavy (118 files) with a React-based UI layer (39 TSX files). The package/ directory contains the actual distributable CLI with vendored binaries for audio capture and ripgrep across platforms.
How To Use It
Setup: The repo has no root package.json; the main one is at Leonxlnx-claude-code/package.json. There is a bun.lock in the package/ directory, suggesting Bun is the package manager for the distributed CLI. The root has only .env.example and .gitignore.
Configuration: Copy .env.example to .env and set the provider key you plan to use. Supported variables include ANTHROPICAPIKEY, OPENAIAPIKEY, GEMINIAPIKEY, GROQAPIKEY, OPENROUTERAPIKEY, COPILOTTOKEN, and ZAIAPIKEY. Ollama requires a running local server with at least one pulled model.
Running it: The launcher scripts at the repo root are the entry points. The exact invocation is not documented in the README excerpt, but the pattern is:
From the Leonxlnx-claude-code/ directory
./claw-dev.sh # Linux/macOS ./claw-dev.cmd # Windows CMD ./claw-dev.ps1 # Windows PowerShell
Node.js 22+ is required.
Real-World Use
A team standardizing on one coding assistant CLI while allowing individual provider preferences: one developer uses ANTHROPICAPIKEY for direct Anthropic access, another uses OPENROUTERAPI_KEY with a custom model ID like anthropic/claude-sonnet-4, and a third runs fully local inference via Ollama. All three invoke the same claw-dev command and get the same terminal experience.
Code Health & Issues
Med - No test files detected - the codebase has 200 files with zero test coverage, leaving the bridge communication and proxy logic untested. Med - No CI/CD pipeline - no .github/ or CI config exists, so there is no automated build or test gate. Low - No lockfile at the source level - Leonxlnx-claude-code/package.json has no accompanying lockfile, so source builds are not reproducible (the package/ directory does have bun.lock). Low - Fork with rebranding risk - the repo is a fork of ahmedmoh1093-wq/claw-dev and includes patch-branding.js and claude-gemini-launcher.js, suggesting a custom fork of Claude Code. Upstream compatibility is a maintenance concern. Low - Missing source file - the README references src/anthropicCompatProxy.ts, but the file listing does not include it; the proxy may live elsewhere or be incomplete.
The Bottom Line
This is a practical multi-provider launcher for Claude Code that solves a real workflow problem for teams with mixed AI provider preferences. It is not production-grade yet: no tests, no CI, and it depends on a vendored third-party binary. Suitable for individual developers or small teams who want provider flexibility without maintaining multiple CLIs, but not ready for enterprise rollout without adding test coverage and a build pipeline.