AI Native Landscape

openclaude

Tracked

An open-source coding-agent CLI for cloud and local model providers — OpenAI-compatible APIs, Gemini, GitHub Models, Codex OAuth, Ollama and more — keeping one terminal-first workflow with tools, agents, MCP, and slash commands.

AuthorTwigpineOpen Sourced2026-04-01Last CommitUnknown

Overview

OpenClaude is an open-source coding-agent CLI that runs anywhere and uses anything. It talks to OpenAI-compatible APIs, Gemini (API key or Vertex AI), GitHub Models, Codex and Codex OAuth, Ollama, and a long list of partner gateways, while keeping a single terminal-first workflow: prompts, bash and file tools, grep and glob, subagents, tasks, MCP, slash commands, and streaming output. A bundled VS Code extension adds launch integration.

Key Features

  • One CLI across cloud APIs and local model backends, with guided provider setup and saved profiles via /provider
  • Coding-agent toolset in one place: bash, file tools, grep/glob, agents, tasks, MCP, and web tools with pluggable search backends
  • Session management: resume or fork conversations by session ID, a /sessions manager, and deep search across conversation contents with a background model
  • Background sessions: run long non-interactive prompts detached (--bg), then ps, logs, and kill them
  • Fastest paths for common setups: OpenAI via two env vars, local Ollama, GitHub Models onboarding via /onboard-github
  • Its own configuration namespace (~/.openclaude) — no Claude Code install or .claude directory required

Use Cases

  • Running a terminal coding agent against whichever provider you already pay for or run locally
  • Switching between cloud models and local Ollama models without changing tools or workflows
  • Kicking off long refactors or test-fix loops as detached background sessions
  • Reusing MCP servers, custom slash commands, and subagent workflows across providers

Technical Details

  • TypeScript CLI distributed as @twigpine/openclaude on npm (Node 24+ required); a Bun toolchain is only needed for source builds, and the repo is also mirrored to an alternate git host
  • Provider abstraction covers 20+ backends: any OpenAI-compatible /v1 server, Gemini API key and Vertex AI ADC, GitHub Models OAuth onboarding, Codex OAuth (browser sign-in with secure credential storage), and Ollama; /provider saves profiles in .openclaude-profile.json and per-provider env vars are mapped automatically
  • Background sessions run as local child processes with no daemon or network service; state and logs live under ~/.openclaude/bg-sessions/, with explicit terminal-state recording (exited/failed/killed/stale)
  • Conversation forking (--fork-session) branches the transcript into a new session ID without touching the working tree; deep session search sends bounded transcript excerpts (first/last 50 messages, 2,000 chars) to a background model
  • Keeps its own config in ~/.openclaude and ~/.openclaude.json and deliberately does not read ~/.claude or CLAUDE_CONFIG_DIR