What one reply looks like
| Part | What you see |
|---|---|
| Reasoning | Folded. Click to read how Lumi planned the scene. |
| Story | The prose itself, in your chat as usual. |
| Choices | Three next moves. Click one to put it in your message box. |
| Recap | Folded one-line summary of what just happened. |
| Note | A short aside from Lumi to you. |
1. Install the extension
- In SillyTavern, open Extensions (the puzzle-piece icon at the top).
- Click Install extension.
- Paste the address below and click Install.
- Reload the page.
https://github.com/tavernapi/sillytavern-lumiYou will find TavernAPI Preset Runtime in the Extensions panel. Updates come through Manage extensions → Update.
2. Import the preset
- Download TavernAPI_StoryEngine.json with the button above.
- Open AI Response Configuration (the sliders icon, top left).
- Click the Import preset icon next to the preset dropdown and pick the file.
- Select TavernAPI_StoryEngine in the dropdown.
The preset already sets temperature, length and streaming. You don't need to change any sliders.
3. Connect a Gemini model
Open API Connections (the plug icon) and set API to Chat Completion. Then pick one:
With TavernAPI
| Field | Value |
|---|---|
| Chat Completion Source | Custom (OpenAI-compatible) |
| Custom Endpoint (Base URL) | https://tavernapi.com/v1 |
| Custom API Key | your key from the TavernAPI console |
| Model | gemini-4-argon (best) or gemini-3.8-flash (cheaper) |
With your own Google key
Choose Google AI Studio (or Google Vertex AI) as the source, paste your key and pick a Gemini model.
Click Connect.
4. Two switches to check
- AI Response Configuration → Enable function calling must be on. Lumi writes her replies through it.
- API Connections → Prompt Post-Processing should be None.
If either is wrong, the extension shows a red notice in its settings.
5. Check that it works
Open Extensions → TavernAPI Preset Runtime. The status next to the title should say Connected (or Ready). Start a chat and send a message: the reply should arrive with the panels above. Lumi thinks before she writes, so the first reply can take a minute.
Make it yours
Everything in the preset list is a switch. Entries with a 🔒 have fixed text, but you can still turn them off.
| You want | Do this |
|---|---|
| Longer replies | Turn on 📝 Word count. |
| No recap | Turn off 🔒 Recap. Older turns are then sent in full. |
| Lumi stops speaking for you | Turn on ⚙️ No puppeting me. |
| Clicking a choice sends it at once | Extensions → TavernAPI Preset Runtime → Choice click: Send. |
| Light panels on a light theme | Panel theme: Auto follows your chat theme; Dark or Light forces one. |
Mature content is off by default. The ❤️ entries turn it on, and the 📖 Read me first entry in the preset explains the rules.
Troubleshooting
| What happens | Fix |
|---|---|
| The reply is one block of text, no panels | The extension is off, or Advanced → Runtime is off in its settings. |
| No reasoning or choices | Enable function calling is off. |
| Status says Offline | The extension can't reach tavernapi.com. It keeps working with its saved copy; check your connection and click ↻. |
| Replies are short | Turn on 📝 Word count, or raise Max response length. |
| An error mentioning tools | Your model or source doesn't support function calling. Use a Gemini model. |