Ollama Flutter GUI is a Flutter chat surface for local Ollama models. You run it against localhost:11434, load the models already pulled on your machine, and keep the chat path on your own Ollama runtime.
It is intentionally small: pick an endpoint, refresh local models, chat through Ollama's /api/chat endpoint, attach a small text file when the prompt needs context, and copy model output from the chat.
Ollama Flutter GUI talks to your local Ollama server. It does not include a hosted model backend, account flow, telemetry path, or cloud fallback.
| Runtime path | Where it runs |
|---|---|
| Flutter UI | Browser |
| Model discovery | GET /api/tags on your Ollama endpoint |
| Chat | POST /api/chat with stream: false |
| File prompt context | In browser memory, then sent to your configured Ollama endpoint |
| Chat history | In app state for the current session |
| Surface | Runtime scope | Status |
|---|---|---|
| Chat | Prompt a local Ollama model and keep prior turns in the request history. | Ready. |
| Model picker | Load models from /api/tags and switch between local tags. |
Ready. |
| Endpoint control | Point the UI at another Ollama endpoint, then refresh models. | Ready. |
| File prompts | Attach one UTF-8 text file up to 1 MB and send it as prompt context. | Ready. |
| Web build | Build as a Flutter Web app. | Covered by CI. |
Requirements: Flutter stable, Dart 3.5+, and Ollama running locally.
git clone https://github.com/gabrimatic/ollama_flutter_gui.git
cd ollama_flutter_gui
flutter pub get
flutter run -d chromeStart Ollama and pull a model:
ollama pull llama3.1
ollama serveDefault endpoint: http://localhost:11434.
- Local model discovery: refreshes the model picker from Ollama's
/api/tagsendpoint. - Chat history: sends previous turns to
/api/chatso the model receives conversation context. - Endpoint switching: edits the Ollama base URL from the app instead of changing source code.
- File prompt context: attaches one small text file and wraps it clearly in the prompt.
- Connection feedback: shows reachable, empty-model, request, and error states in the UI.
- Copy output: tap a message to copy it.
- Flutter Web first: the main path is
flutter run -d chromeandflutter build web --release.
Run the normal checks:
flutter pub get
flutter analyze
flutter test
flutter build web --releaseCI runs the same analyze, test, and web build path on pushes and pull requests.
| Path | Purpose |
|---|---|
lib/main.dart |
App bootstrap and theme |
lib/chat_screen.dart |
Chat UI, endpoint control, model picker, composer |
lib/chat_state.dart |
Riverpod state, file prompt handling, request flow |
lib/ollama_service.dart |
Ollama API client for /api/tags and /api/chat |
lib/chat_message.dart |
Message model and Markdown-rendered chat bubble |
test/chat_state_test.dart |
Local HTTP contract tests for model discovery, chat, and files |
Default values live in lib/consts.dart:
| Setting | Default | Notes |
|---|---|---|
kDefaultOllamaBaseUrl |
http://localhost:11434 |
Change in the UI or source. |
kDefaultAiModel |
llama3.1 |
Used until /api/tags returns a local model. |
kMaxUploadBytes |
1048576 |
One MB per attached file. |
- Streaming is not enabled yet; requests use
stream: false. - Files are sent as prompt text, not as binary uploads.
- Chat history is session-local and clears when the app reloads.
- Browser CORS still depends on how your Ollama server is exposed. If you serve the UI from a LAN or Tailscale address, allow that origin in Ollama before expecting browser requests to work.
Ollama API · Flutter docs · Soroush · GitHub
MIT License. Details: LICENSE.