OHarness vs Claude Code
Choose Claude Code if you work with Claude models and want the best-supported tool for them, with an OS sandbox and usage included in a Claude plan. Choose OHarness if you want the same kind of agent on any model — OpenAI, Gemini, DeepSeek or a local one — switching mid-session without losing the conversation.
Claude Code’s side checked against its own documentation on . Sources are listed below.
Ta strona jest napisana po angielsku. Terminy, które wyjaśnia, są tymi, których szuka się po angielsku, a przetłumaczony termin fachowy to już inny termin.
Side by side
| OHarness | Claude Code | |
|---|---|---|
| Form factor | Terminal UI, browser UI, desktop app (macOS) and a headless runner — one engine | Terminal CLI, VS Code and JetBrains extensions, desktop app (macOS, Windows, Linux beta), web, the Claude mobile apps, Slack and Chrome |
| Licence and source | Free to use; npm packages MIT. Source repository not public yet | Proprietary, under Anthropic's Commercial Terms; the public repository holds plugins and examples, not the source |
| Models | Anthropic, OpenAI and Gemini natively; any OpenAI-compatible endpoint; 30+ catalogued providers | Claude models only — through the Anthropic API, Amazon Bedrock, Google Cloud or Microsoft Foundry |
| Local models | Ollama, LM Studio and vLLM built in, no key | Not supported: Anthropic does not support routing Claude Code to non-Claude models |
| Switching mid-session | /model mid-session; history is vendor-neutral, nothing converted | /model switches between Claude models; opusplan plans with Opus and executes with Sonnet |
| MCP | Yes — stdio and Streamable HTTP, with OAuth for remote servers | Yes — stdio, HTTP and WebSocket, OAuth for remote servers, in local, project and user scopes |
| Permissions | Five modes plus allow/ask/deny rules in user, project and local scopes; deny beats bypass | Six modes, from default to bypassPermissions; deny, then ask, then allow; user, project, local and managed scopes |
| Sandbox | No OS sandbox — a policy gate on tool arguments. Use a container for confinement | OS-enforced for shell commands: Seatbelt on macOS, bubblewrap on Linux and WSL2, with a network proxy; turned on with /sandbox |
| Subagents | Yes — isolated contexts, run in parallel | Yes — Explore, Plan and General-purpose built in; custom ones as Markdown files |
| Headless | oharness-headless, JSONL on stdout | claude -p with text, JSON or stream-JSON output; GitHub Actions; the Agent SDK for Python and TypeScript |
| Price | Free with your own key or a local model; optional hosted plan | Included in Claude Pro, Max, Team and Enterprise plans, or billed per API token; not on the free plan |
Where Claude Code is stronger
- It is built by the company that trains the models it drives, and it is the best-supported way to use them.
- A real OS sandbox for shell commands, with network access behind a proxy. OHarness has a policy gate, not confinement.
- It runs in more places: IDE extensions, a desktop app on Windows and Linux, the web, mobile, Slack, and cloud sessions that keep working when the laptop is closed.
- Checkpoints with
/rewind, managed settings an organisation can enforce, and a plugin marketplace. - Usage comes with a Claude subscription, so there is no per-token bill to watch.
Where OHarness is stronger
- Any model: Anthropic, OpenAI, Gemini, DeepSeek, Kimi, GLM or a local Ollama model — and more than one of them in the same session.
- A mid-session switch across vendors converts nothing; Claude Code's
/modelmoves between Claude models only. - Ollama, LM Studio and vLLM are built in and need no key; Claude Code does not support local models.
- Free to use with your own provider key; no subscription is required.
Which to choose
Claude Code, if you use Claude models and want the tool built around them: an OS sandbox, IDE, desktop, web and mobile surfaces, cloud sessions, and usage included in a Claude plan.
OHarness, if you want Claude alongside other vendors or a local model in one session, on your own key, without a subscription — or you want to stop depending on a single vendor's pricing and capacity.
Common questions
Is OHarness a clone of Claude Code?
No. It follows the same architecture — an agent loop, tools, permissions, subagents and MCP in one layer — but it is an independent project, not affiliated with Anthropic, built so that the model is a replaceable backend.
Can Claude Code use DeepSeek, Gemini or a local model?
Not officially: Anthropic's documentation says it does not support routing Claude Code to non-Claude models. Some vendors, DeepSeek and Ollama among them, publish Anthropic-compatible endpoints that work with it anyway, with limits — one vendor per session, and features such as prompt caching may be ignored. The guide on using Claude Code with other models covers the setup and the trade-offs.
Can I use Claude models in OHarness?
Yes. The anthropic provider speaks the Anthropic Messages API natively with an ANTHROPIC_API_KEY, so the same Claude models are available — alongside OpenAI, Gemini, DeepSeek and local models in the same session.
Is OHarness better than Claude Code?
For Claude models alone, Claude Code is the better-supported tool and has an OS sandbox OHarness lacks. OHarness is the better fit when you want more than one vendor, a local model, or no subscription — it is free to use with your own key.
Sources
- Claude Code overview
- Model configuration
- LLM gateways
- Permissions
- Sandboxing
- MCP
- Subagents
- Headless mode
- Setup and plans
- Repository and licence
Spotted something out of date? Tell us and it will be corrected. More comparisons are on the guides page, and Claude Code alternatives covers what to check before switching.