What to actually compare
The criteria that matter
| Criterion | Why it matters | What to look for |
|---|---|---|
| Price | You don't want a per-seat subscription for a coding agent | Free to use; you only pay your own model usage |
| Open source | Auditability + self-hosting + no lock-in | A real OSS license (e.g. MIT), code you can read |
| Local execution | Reads your real files, no sandbox bridge to break | Runs as a normal process on your machine |
| Bring your own model | Not tied to one provider's cost/limits/outages | Provider, base URL, key, model are settings |
Why 'free + open source' matters here
A coding agent is, at its core, a model plus a loop that runs tools on your files. The valuable, differentiated part — the model — you're already paying for via your own key. So paying a second subscription just for the agent shell is often unnecessary. Free and open source also means you can read exactly what it does with your code, and self-host it if you want. That's hard to beat on trust.
Why 'local + bring your own model' matters
Two of the most common Claude Code complaints — rate limits/quota and being stuck on one provider — both come from being bound to a single account. An agent that lets you bring your own model routes around that: when one provider is capped or down, you switch. And running locally means it reads your real filesystem directly, instead of a connector that can be "connected but empty."
Where Wuwei lands
Free
No subscription for the agent — you only pay your own model usage.
Open source (MIT)
Read the code, self-host, no black box.
Local
Runs as a normal desktop app and reads your real files directly.
Bring your own model
Claude, OpenAI-compatible endpoints, GLM / Kimi / DeepSeek — switch in a click.
FAQ
Is there a free alternative to Claude Code?
Yes. Wuwei is a free, open-source (MIT) coding agent — you don't pay a subscription for the agent itself, only your own model usage. It runs locally and lets you bring any model.
Is there an open-source Claude Code alternative?
Yes. Claude Code itself isn't open source; Wuwei is MIT-licensed, so you can read the code and self-host it. It runs locally and is model-agnostic.
Can an alternative use the same models as Claude Code?
Yes — a bring-your-own-model agent can point at Claude, OpenAI-compatible endpoints, GLM, DeepSeek, or a local model. You're not restricted to one provider, so you can pick per task.