How we picked
We include a tool when its docs or shipped config show local model servers, such as Ollama, LM Studio, llama.cpp or vLLM. Where we found nothing, the tool is left out, which does not prove it cannot work. Codex CLI is one case: only third-party guides mention local use.
Things to check before you pick
- Depth of documentation varies. OpenCode documents Ollama, LM Studio and llama.cpp. Forge shows local providers in its config, but its README does not explain them. Qwen Code's README names Ollama and vLLM, while its auth docs mention only a custom provider.
- Some local routes still need an account. GitHub Copilot CLI accepts Ollama and vLLM through bring-your-own-key, but it needs a Copilot subscription. Factory Droid allows local endpoints but requires a Factory sign-in.
- Hardware matters. Cline's docs recommend compact prompts and strong hardware for local models. mini-SWE-agent has a vLLM example, but Ollama is only a config value.
- Codewhale detects a running Ollama automatically, and Open Interpreter targets cheap open-weight models.