LLM Chat¶
LLM Chat is a minimal, single-model chat surface at /admin/llm-chat. It is intentionally simpler than the Personal Assistant: no conversation history, no attachments, no action blocks. Use it for quick prompts when you want a clean test of a specific model.
1. When to use LLM Chat vs Personal Assistant¶
| Feature | LLM Chat | Personal Assistant |
|---|---|---|
| Where | /admin/llm-chat (full page) |
Sidebar drawer on every page |
| Conversation history | No | Yes |
| File attachments | No | Yes |
| Voice input | No | Yes |
| Model picker | One model per session | Speed tier (Fast / Balanced / Powerful) |
| Action blocks / tool calls | No | Yes |
| Tytus AIL gateway | No | Yes (when pod connected) |
| Use case | Quick prompt against one specific model | Daily work, in-app help, agent interactions |
2. Send a prompt¶
- Open Admin → LLM Chat in the sidebar.
- Pick a model from the dropdown. The list comes from the platform's available chat models.
- Type your prompt in the input.
- Press Send.
The response renders as plain text with markdown formatting. No streaming bubbles, no thinking accordion.
3. Switch models¶
Change the model dropdown at any time. Switching wipes the on-page response — there is no per-model history to preserve.
4. Errors¶
If the model returns an error (rate limit, content filter, network), an inline message replaces the response area. The service auto-resets on the next prompt. No retry-with-different-model fallback.
5. What's not here (by design)¶
LLM Chat is not the place for:
- Multi-turn conversations — use Personal Assistant.
- File analysis — use Personal Assistant.
- Agent workflows — use Projects + Studio Tools.
- Tytus AIL gateway access — use Personal Assistant with your pod connected, or call your pod's OPENAI_BASE_URL directly.
For anything beyond a quick prompt, Personal Assistant is the right surface.
Mirrored from traylinx-web:docs/user-manuals/llm_chat.md. Edit the source in the traylinx-web repo — changes here are overwritten by the sync script.