IsuruMaduranga/one-code

Claude Code with the model slot left open. The full workflow, open source, on any model or provider

Summary

One Code is an open-source TypeScript terminal coding agent that recreates the Claude Code workflow while leaving model and provider selection to the user. Built on pi, it includes subagents, Git worktrees, planning, permissions, auto mode, and context management, and reads existing CLAUDE.md and .claude/ configuration directly. Engineers can use one model for the main task and assign different models or providers to exploration, testing, and review.

What it is useful for

One Code is useful for teams that want a Claude Code-style terminal workflow without tying every task to one model. Supported choices include Anthropic, OpenAI, Gemini, OpenRouter, and local models. The main model can be changed with /model, while /subagent <provider/model-id> sets the default for delegated work. For example, an engineer could ask subagents to investigate authentication code and its tests, then have the main session implement the fix based on their findings. /agents shows delegated-agent progress.

It also helps teams reuse an existing Claude Code setup. One Code reads CLAUDE.md, falling back to AGENTS.md, along with .claude/skills/, commands, agent definitions, permission rules, plugins, and .mcp.json configuration. There is no import step. For parallel code changes, agents can work in separate Git worktrees, while plan mode lets the agent investigate and write a plan before implementation is approved.

The project is not a complete replacement for Claude Code: features that depend on Claude Code’s hosted or desktop services are explicitly outside its scope. End-to-end testing focuses on Anthropic, OpenAI, and OpenRouter, while other providers have less coverage. Permissions provide application-level controls rather than operating-system isolation, so the README recommends running One Code in a container when stronger isolation is required.

How engineers can use it

Install Node.js 22.19 or later, then install and launch One Code from the target repository:

npm install -g @one-ai/one-code
cd your-project
onecode

Inside the session, run /login to connect a provider and /model to select a model. API keys can instead be supplied through documented variables such as ANTHROPIC_API_KEY, OPENAI_API_KEY, or OPENROUTER_API_KEY.

For a typical review workflow, choose the main model, configure the subagent model, ask the agents to inspect a specific area, and review the proposed plan before allowing edits. Use onecode doctor or /doctor report to check credentials, model assignments, detected configuration, and missing programs.

For scripted repository analysis, the README also documents:

onecode -p "Explain how authentication works in this repository"
onecode --mode json