A browser side panel for chatting with local and self-hosted LLMs. Your chats, files and embeddings stay on your machine.
Docs · Provider setup · Privacy · Issues
- Install from the Chrome Web Store (Chrome, Edge, Brave) or Firefox Add-ons.
- Start a model server — Ollama, LM Studio or llama.cpp. For Ollama,
olcdoes the setup for you. - Open the side panel, pick a model, chat.
- Streaming chat in the side panel, with stop, edit-in-place, branching, tags, export, print and backup.
- Providers: Ollama, LM Studio and llama.cpp built in; vLLM, LocalAI, KoboldCPP or any OpenAI-compatible server, and Anthropic, as custom providers.
- Local RAG over your files: hybrid keyword + vector search, with offline embeddings bundled in the extension.
- Browser context: attach tabs, selected text, files or images; tool-capable models can read the current tab and open tabs themselves.
- Web search (optional) through SearXNG, Brave or Tavily.
- Vision for models that support images.
- Selection button: send highlighted text on any page straight to chat.
- Browser agent (experimental, off by default): the model can open sites, search, click and fill forms, asking you before anything consequential.
- Privacy screen: see everything stored, back it up, or wipe it. No telemetry.
| Feature | Chrome / Edge / Brave | Firefox |
|---|---|---|
| Chat, providers, history | ✅ | ✅ |
| Local RAG and offline embeddings | ✅ | ✅ |
| Images, web search, tab tools | ✅ | ✅ |
| Selection button | ✅ | ✅ |
| Tab groups | ✅ | optional permission |
| Browser agent (experimental) | ✅ | ❌ |
| Minimum version | 116 | 114 |
The agent drives pages through Chrome's debugger API, which Firefox does not offer, so Firefox builds leave it out entirely.
| Provider | Default endpoint |
|---|---|
| Ollama | http://localhost:11434 |
| LM Studio | http://localhost:1234/v1 |
llama.cpp (llama-server) |
http://localhost:8000/v1 |
| OpenAI-compatible | your URL |
| Anthropic | https://api.anthropic.com/v1 |
Models with no saved provider mapping route to Ollama. See the capability matrix.
olc starts or reuses Ollama with extension access already configured, so you skip setting OLLAMA_ORIGINS by hand. It needs Node.js 22.12+.
curl -fsSL https://ollamaclient.in/olc.sh | sh # macOS / Linuxirm https://ollamaclient.in/olc.ps1 | iex # WindowsThese run the install script straight from the site without checking it first. To pin a release and verify its checksum before anything runs, use the verified install.
Then:
olc # Ollama with extension access
olc --lan # also reachable from your trusted network
olc -b codex # Codex as an OpenAI-compatible provider on :8083
olc -b opencode # OpenCode on :8084
olc -b fm # Apple's on-device model on :8085 (macOS 27)
olc -b laya # Laya decision API on 127.0.0.1:8086
olc -b searxng # local web search on 127.0.0.1:8080
olc list # running Ollama, proxy, and Docker services
olc --helpAdd the Codex, OpenCode or Apple endpoints in the extension as a custom OpenAI-compatible provider at http://127.0.0.1:<port>/v1.
Check the installed olc version
Runolc --version and compare it with the release you expect.Run SearXNG through olc. Docker Desktop or Docker Engine with the Compose v2 plugin must be installed and running; olc creates the local config and starts SearXNG with Valkey without a repository clone or downloaded setup files:
olc -b searxng
olc -b searxng status
olc list
olc -b searxng stop # pause; preserve containers and data
olc -b searxng rm # remove containers; preserve config and data
olc -b searxng rm --purge-data # delete SearXNG data and managed configSet http://localhost:8080 in Settings → Knowledge & web → Web search. If port 8080 is occupied, start it with olc -b searxng --port 18080 and use http://localhost:18080. SearXNG listens on 127.0.0.1 only. stop pauses its containers; rm removes them while preserving config and data volumes. Add --purge-data to rm to delete those too. If the Compose file was customized, purge leaves it, .env, and settings (including the SearXNG secret) in place so user-added services are not orphaned; those services and volumes are left untouched. The CLI prints the retained config files. Manage or stop added services with the customized Compose file, then inspect the directory before cleanup. Remove only listed OLC config files if they are no longer needed, and preserve all other files and directories because they may contain user data. Fresh olc installs use a pinned SearXNG image; existing .env version selections are preserved. Search results depend on upstream engines, which may rate-limit or challenge requests. Laya runs locally at http://127.0.0.1:8086/v1/systemone (use olc -b laya --port <port> if needed). To change a Laya port, remove the container with olc -b laya rm, then start it with the new --port; the model cache is preserved. Brave and Tavily need only an API key.
Local providers keep everything on your machine. A remote provider sees what you send it, and a remote search provider sees your queries. Chat history, files and embeddings are stored locally. Diagnostics are content-free and never uploaded. Don't expose a model server to the internet without authentication.
git clone https://github.com/Shishir435/ollama-client.git
cd ollama-client
pnpm install
pnpm dev # Chrome
pnpm dev:firefox # FirefoxRun pnpm verify before a PR. See CONTRIBUTING.md, AGENTS.md for architecture and conventions, and tools/README.md for commands.