Lucy AI/Docs
GitHubOpen app

Getting started

  • Introduction
  • Quick start
  • Desktop app

Using Lucy

  • Chat & models
  • Memory
  • Personas
  • Connectors
  • Workflows
  • Voice
  • Security & 2FA
  • Themes & account

Developers

  • Architecture
  • Self-hosting
  • Embedding Lucy
  • HTTP API
  • CLI
  • MCP server
  • Contributing

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.

ProviderModels
OpenAIGPT-4o · GPT-4o Mini · GPT-3.5 Turbo
Anthropic (Claude)Claude Opus 4.8 · Claude Sonnet 4.6 · Claude Haiku 4.5
Google (Gemini)Gemini 2.0 Flash · Gemini 1.5 Pro
DeepSeekDeepSeek V3 (Chat) · DeepSeek R1 (Reasoner)
GroqLlama 3.3 70B · Llama 3.1 8B Instant · Gemma 2 9B — ultra-fast LPU inference
MistralMistral Large · Mistral Small
xAIGrok 2 · Grok Beta
OpenRouterAuto-route, plus 200+ models behind one key (e.g. Claude 3.5 Sonnet, Llama 3.3 70B)
LocalAny Ollama or LM Studio model you've pulled — auto-discovered, no API key

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/<provider>.ts.

Message tools

  • Edit (pencil icon on your messages): rewrites the message and re-sends from that point. Cmd/Ctrl+Enter saves, Escape 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:

CommandEffect
/remember <fact>Save a fact to Lucy's memory
/global <fact>Save a fact visible across all projects
/forget <topic>Remove matching memories
/memoriesList what Lucy knows
/incognitoToggle no-capture mode for this conversation
/newStart a fresh conversation
/helpShow all commands

Keyboard shortcuts

ShortcutAction
Cmd/Ctrl + KFocus conversation search
Cmd/Ctrl + Shift + NNew chat
EscapeClose 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.

Desktop appMemory