Chat & models
Choosing a model
The model selector lives in the chat input bar. Cloud providers are always listed; the Local group appears when Lucy detects a running Ollama (localhost:11434) or LM Studio (localhost:1234) server. Use the refresh button next to the selector to re-probe.
You can switch models mid-conversation - the full history goes to the new model, so you can draft with a fast model and polish with a stronger one, or compare answers side by side.
Providers & models
Lucy reaches every major provider through one interface. Add your own key in Settings → Providers (kept client-side in standalone mode, AES-256-GCM encrypted server-side in connected mode) - or run fully local with no key at all.
| Provider | Models |
|---|---|
| OpenAI | GPT-6.1 Sol (default for new users) · GPT-6 Astra · GPT-6.1 Terra · GPT-6 Luna |
| Anthropic (Claude) | Claude Fable 5.2 · Claude Opus 5.5 · Claude Sonnet 5.5 · Claude Haiku 5.5 |
| Google (Gemini) | Gemini 3.6 Flash · Gemini 3.8 Flash · Gemini 3.5 Flash-Lite · Gemini 4 Argon |
| DeepSeek | DeepSeek V4 Pro · DeepSeek V4.1 Flash |
| Groq | GPT-OSS 120B · GPT-OSS 20B · Llama 3.3 70B · Llama 3.1 8B Instant |
| Mistral | Mistral Large · Mistral Small |
| xAI | Grok 4.7 · Grok 4.6 |
| OpenRouter | Auto-route, plus hundreds of models behind one key |
| Local | Any Ollama or LM Studio model you've pulled - auto-discovered, no API key |
Older models (marked "previous") stay in the list so existing chats and workflows keep working.
Server-level keys (set as env vars) act as a shared fallback when a user hasn't entered their own. Local models appear automatically once their server is running; the cloud catalog is defined in lib/providers/catalog.ts.
Message tools
- Edit (pencil icon on your messages): rewrites the message and re-sends from that point. Cmd/Ctrl+Enter saves, Escape cancels.
- Reply (arrow icon on any message): quotes that message above the box; your answer shows the quote on top, and a click on the quote jumps to the original. Tag @lucy or another AI in the reply to ask about that message. Escape or × cancels.
- Regenerate (on Lucy's replies): re-runs the preceding prompt.
- Copy any message; code blocks have their own copy button with language labels and line numbers.
- Read aloud (speaker icon): speaks the reply using your configured voice.
- Token counts appear per message on hover and as a conversation total in the header.
- Export a conversation as Markdown or JSON from the ⋯ menu.
Slash commands
Type / in the input for autocomplete:
| Command | Effect |
|---|---|
/remember <fact> | Save a fact to Lucy's memory (only for you) |
/team <fact> or /company <fact> | Save a fact for everyone in your workspace (enterprise) |
/global <fact> | Teach Lucy a fact for every user (Lucy admins only) |
/brainstorm | Turn on the Brainstorm skill: Lucy asks one question at a time and ends with a written plan |
/<skill name> | Turn on one of your skills for this chat |
/forget <topic> | Remove matching memories |
/memories | List what Lucy knows |
/incognito | Toggle no-capture mode for this conversation |
/new | Start a fresh conversation |
/help | Show all commands |
Ask other AIs in the chat (enterprise)
Type @ and pick an AI to call it into the chat with your own key for that provider:
| Tag | Answers with |
|---|---|
@grok | Grok 4.7 (xAI) |
@claude | Claude Sonnet 5.5 (Anthropic) |
@gpt | GPT-6.1 Sol (OpenAI) |
@gemini | Gemini 3.6 Flash (Google) |
@deepseek | DeepSeek V4 Pro |
@mistral | Mistral Large |
@llama | Llama 3.3 70B (Groq) |
- One tag: only that AI answers the message, shown with its name and model.
- Two or more tags: a discussion. The AIs answer in turn and react to each other, then Lucy sums up: where they agree, where they differ, and her pick.
- Options (in the bar above the input): 1 to 3 rounds, "Independent first" (in round 1 nobody sees the other answers), and Lucy's summary on or off.
- Guest AIs get the chat, not your personas, memories or skill instructions.
- Tags are coloured: AIs purple, teammates blue.
If a key is missing, a red banner says which one to add in Settings → Providers; the other AIs still answer.
Shared chats (enterprise)
In a team workspace you can work in one chat together:
- Share in the chat toolbar shares the chat with everyone in the team or with the people you pick. Typing
@and a teammate's name also adds them. - Shared chats appear under Shared with you in the sidebar. Every message shows who wrote it, and each person's question is answered with their own settings and keys.
- Live: when two people have the same chat open, a teammate's message appears at once, their answer is written on your screen word by word, and a line above the input shows who is here and who is typing. You can both ask at the same time.
- Only the owner renames, deletes or changes who the chat is shared with.
Your chat list (enterprise)
- Newest on top in every section; a teammate's new message moves the chat up at once.
- Pin (hover a chat, pin icon): pinned chats sit in Pinned at the top. Pins are yours only.
- Folders for your own Lucy chats: the folder icon next to New chat makes a folder; the folder icon on a chat moves it into one. Click a folder's name to open or close it; the pencil renames or removes it (its chats go back to the list). Teammate chats, groups and shared chats keep their own sections.
- Long chats load in pages: a chat opens on its last 50 messages; scroll up for older ones.
- Reload keeps your place: the open chat is in the address, so refreshing the page brings you back to it.
- Lucy was updated: after a new version goes live, open tabs offer a reload.
Chats with teammates (bubbles, reactions, Seen, notifications, calls) are in Members.
Keyboard shortcuts
| Shortcut | Action |
|---|---|
Cmd/Ctrl + K | Focus conversation search |
Cmd/Ctrl + Shift + N | New chat |
Escape | Close mobile sidebar |
Tool use
If you have connectors installed, Lucy can call their tools mid-reply. Tool calls appear as colored chips above the streaming message - gray while running, green on success, red on failure.