Gitlawb/openclaude

OpenClaude – an open‑source, terminal‑first coding‑agent CLI

What it is

• A Node.js‑based command‑line interface that lets you talk to a wide range of LLM providers (OpenAI‑compatible APIs, Gemini, GitHub Models, Ollama, Atomic Chat, etc.) from a single terminal window.
• It bundles a set of coding‑agent capabilities: bash execution, file read/write, grep/glob, multi‑step tool calling, and a “repo map” that injects a structural view of the current codebase into the prompt.
• A companion pixel‑art “buddy” can be summoned for a bit of terminal flair.

Key features (as described in the README)

• Unified provider handling – /provider walks you through setup and stores profiles in ~/.openclaude-profile.json. Supports dozens of cloud gateways (OpenAI, Gemini, Fireworks AI, etc.) and local back‑ends like Ollama.
• Tool‑driven workflows – built‑in commands for bash, file editing, globbing, and custom agents; the CLI streams token‑by‑token output and can call tools repeatedly in a single conversation.
• Background sessions – run non‑interactive prompts detached from the terminal (openclaude --bg …), list them with ps, view logs, or kill them. All session data lives under ~/.openclaude/bg-sessions/.
• Repo map – an automatically generated graph of the repository (PageRank‑ranked) that can be injected into context via the REPO_MAP flag (/repomap to view).
• VS Code extension – a bundled extension lets you launch OpenClaude from the editor and provides theme support.
• Companion buddy – optional true‑color pixel‑art character that reacts to each entered message (/buddy).

Typical use cases

• Quickly ask an LLM to generate, refactor, or debug code without leaving the terminal.
• Run multi‑step automation where the model decides to invoke shell commands or edit files.
• Use a local model (Ollama) for privacy‑sensitive work, or switch to a cloud provider for larger models.
• Keep long‑running tasks (e.g., test fixing) in the background while you continue other work.
• Leverage the repo‑map to give the model awareness of a large codebase.

Installation & quick start

# Requires Node ≥22
npm install -g @gitlawb/openclaude@latest   # or via AUR on Arch
openclaude                                 # launch the REPL

• Run /provider to configure a backend (API key, base URL, default model).
• Use /onboard‑github for GitHub Models, or set environment variables such as OPENAI_API_KEY and OPENAI_MODEL for OpenAI.
• Resume or fork conversations with --resume <id> / --continue and --fork-session.
• Background jobs: openclaude --bg "fix failing tests".

Supported providers (excerpt)

Provider	How to configure	Notes
OpenAI‑compatible (OpenAI, OpenRouter, DeepSeek, Groq, LM Studio, etc.)	/provider or env vars	Uses standard /v1 chat endpoint
Gemini	/provider or env vars	API‑key only
Ollama	/provider or env vars	Local inference, no API key
GitHub Models	/onboard‑github	Interactive onboarding
Atomic Chat, Bedrock, Vertex, etc.	/provider or env vars	Various cloud gateways
Many partner gateways (AI/ML API, Novita AI, Concentrate, Exa, etc.)	/provider	Auto‑discovers models from each service

Ecosystem & community

• Discussions & Discord – links in the badge bar for GitHub Discussions and a public Discord server.
• VS Code extension – launch OpenClaude from the editor and get theme integration.
• Partner integrations – listed partners (GitLawb, Bankr.bot, Atomic Chat, Xiaomi MiMo, Atlas Cloud, etc.) provide API keys or gateways that OpenClaude can route to.
• Security & licensing – MIT‑licensed, with a SECURITY.md file and a formal security policy.

Maturity

• Actively maintained (badge shows recent daily/monthly Trendshift stats, CI badge for PR checks, npm download badge).
• Supports a broad set of providers and has advanced features like repo‑map, background sessions, and gRPC server (see docs).
• The README notes provider‑specific quirks (e.g., Anthropic features may be missing elsewhere, smaller local models may struggle), indicating real‑world usage feedback.

Bottom line OpenClaude is a genuine, open‑source CLI that brings LLM‑powered coding assistance into the terminal, with flexible backend support, tool‑calling loops, and optional UI niceties. It is well‑documented, extensible, and positioned for developers who want a single, scriptable interface to many LLM services.