OpenClaude — Universal Terminal AI Coding Agent
One CLI for cloud and local model providers, with a full coding-agent workflow.
Evidence shows permission control mechanisms (canUseTool callback, permission modes), but no detailed explanation of default least privilege; user confirmation exists (permission prompts), but default behavior not explicit; data flow transparency: README mentions config storage location, but not detailed data flow; sensitive data handling: security policy and privacy verification script exist, but no encryption details; dependency security: pinned dependencies and overrides, but no vulnerability scan evidence; external effects: WebSearch and WebFetch tools exist, but default network behavior not specified; rollback: no rollback mechanism mentioned; source attribution: README declares derivation from Claude Code, but no specific code provenance.
Self-consistency: README and package.json descriptions align, but some inconsistencies (version number, license declaration); dependency availability: pinned versions, but no availability guarantee; failure messages: error handling code exists, but no user-friendly failure message examples.
Audience and scenarios: README targets developers and non-technical users, provides multiple installation and setup guides; capability boundaries: clearly lists supported features and limitations; trigger precision: commands and tools defined clearly, but no detailed trigger conditions; environment fit: supports multiple OS and Node.js versions, but no test evidence for all environments.
Information architecture: README well-structured with TOC and sections; install notes: multiple installation methods, but no detailed troubleshooting; naming stability: version exists, but no naming convention; examples and FAQ: multiple examples, but no FAQ; known limitations: partially mentioned, not comprehensive; license: license file exists, but declaration unclear (NOASSERTION); versioning changelog: not provided; maintenance responsibility: security policy and community channels, but maintainer not explicit.
Output usability: CLI and SDK provided, output format not detailed; marginal value: multi-provider support, but no comparison with other tools; cost benefit: no performance or cost data.
Claim traceability: README claims lack specific evidence; cross-source corroboration: CI and tests exist, but no independent verification; fact/inference separation: not explicitly distinguished.
- License declared as NOASSERTION, and LICENSE file indicates code derived from Anthropic's proprietary software, posing legal risk.
- No rollback mechanism provided, users cannot revert to previous state on issues.
- Dependency security lacks vulnerability scan evidence, potential risk.
- No changelog provided, users cannot track updates.
What does this agent do, and when should you use it?
OpenClaude is an open-source coding-agent CLI designed to unify cloud and local model providers. It supports OpenAI-compatible APIs, Gemini, GitHub Models, Codex OAuth, Codex, Ollama, Atomic Chat, and more, while maintaining a single terminal-first workflow: prompts, tools, agents, MCP, slash commands, and streaming output. Key features include guided provider setup (/provider), saved profiles, multi-agent routing and step limits, and a bundled VS Code extension. OpenClaude also offers a headless gRPC server for integration into other applications. It originated from the Claude Code codebase but has been substantially modified for multi-provider support, with modifications under the MIT license.
OpenClaude runs as a command-line tool (openclaude) with configurable provider backends. It executes tool-driven coding workflows: bash, file read/write/edit, grep, glob, agents, tasks, MCP, and slash commands. The /provider command guides setup and saves profiles; it can call local services like Ollama or cloud APIs like OpenAI. It supports streaming responses, multi-step tool loops, image inputs (URL or base64), and repo map generation (REPO_MAP flag). Advanced features include background sessions (--bg), resume (--resume), and forking (--fork-session). A pixel-art buddy (/buddy) provides visual feedback. The repo also includes a VS Code extension and a headless gRPC server (src/proto/openclaude.proto).
- Developers who switch between model providers and want a consistent CLI workflow.
- Users running local models with Ollama who need the same tooling as cloud APIs.
- Teams integrating agentic capabilities into CI/CD pipelines or custom UIs via gRPC.
- VS Code users seeking launch integration and theme support through the bundled extension.
- Users who want to leverage GitHub Models or Codex authentication but need a unified entry point.
- Operators running long non-interactive prompts in the background and monitoring logs.
What are this agent's strengths and limitations?
- Unified workflow across many providers, avoiding per-provider tooling.
- Local model support via Ollama without an API key.
- Advanced features like repo map and background sessions.
- Includes VS Code extension and headless gRPC server for integration.
- Active community with Discord and X channels.
- Provider behavior varies; tool quality depends on the chosen model.
- Requires Node.js >=22.0.0, which may be a hurdle for some environments.
- Some providers have limitations (e.g., AI/ML API only surfaces chat-capable models).
- Background sessions lack full terminal reattach.
- Derived from Claude Code; license covers modifications only, with original code under Anthropic's copyright.
How do you install or deploy this agent?
Requires Node.js >=22.0.0. Install globally: npm install -g @gitlawb/openclaude@latest. Arch Linux: paru -S openclaude. If ripgrep is missing, install it system-wide and verify with rg --version.
How do you use this agent?
Start with: openclaude. On first run, use /provider for guided provider setup, saving profiles to .openclaude-profile.json. Example (OpenAI): export CLAUDE_CODE_USE_OPENAI=1; export OPENAI_API_KEY=sk-...; export OPENAI_MODEL=gpt-4o; openclaude. Resume or continue: openclaude --continue or openclaude --resume <session-id>. Background: openclaude --bg "fix tests"; view logs with openclaude logs <name>.
How does this agent compare with similar options?
Compared to Claude Code, OpenClaude is designed for multi-provider support, whereas Claude Code is primarily for Anthropic models.