v0.2.65 · macOS ready

A desktop AI assistant
for your computer — not your browser tab.

One surface for every model. Run Qwen, Gemma, Llama, DeepSeek locally via Ollama — or route to Claude, GPT, Gemini, Grok, Kimi, Qwen, GLM when you need the heavy hitters. Domain-tuned Spaces, real tool calling in a workspace folder, and telemetry that stays on your disk.

↓ Download for macOS ★ View on GitHub
Free · MIT licensed · Works offline with ollama
Alaude
Analyze my CBC panel. HGB 11.2, MCV 68, ferritin 8.
🩺 Health Space claude-opus-4
Low hemoglobin with microcytosis and very low ferritin points to iron-deficiency anemia. Running lab reference check and flagging red flags:
🛠 analyze_lab_result 🛠 red_flag_triage 🛠 score_phq9
Informational only — not a diagnosis. See a clinician.
Works with
🟠 Anthropic Claude
🟢 OpenAI GPT · o-series
🔵 Google Gemini
⚫ xAI Grok
🌙 Moonshot Kimi
☁️ Alibaba Qwen
🔷 Zhipu GLM-5.1
🦙 Ollama (local)
Features

Everything a single-model chat app isn't.

Alaude is the Swiss Army knife between you and every LLM you care about — local or cloud, chat or tool-running, general or domain-tuned.

🦙

Local models via Ollama

One-click install of Qwen 3.6, Gemma 4, Llama 3.2/3.3, DeepSeek R1, and anything else with an Ollama tag. Progress bar, cancel button, and it shows up alongside cloud models in the dropdown.

☁️

Every major cloud provider

Anthropic, OpenAI (GPT-4, o3, o4), Google Gemini, xAI Grok, Moonshot Kimi, Alibaba Qwen, Zhipu GLM-5.1. Keys live in ~/.claude/.credentials.json with mode 0600.

🧭

Domain-tuned Spaces

Seven built-in modes with their own system prompts and quick-actions: General, Health, Finance, Real Estate, Legal, Education, Marketing. Create your own in-app.

🛠️

Real tool calling

Point Alaude at a workspace folder and the model can read_file, write_file, list_directory, run_command, open_in_browser, and start_dev_server. Activity chips show every call live.

📊

OODA loop & self-auditing UX

Every interaction is logged locally. A priority-ordered analyzer runs six diagnose rules — provider error rates, retry patterns, quick-action abandonment, latency outliers — and proposes fixes. Humans still decide.

🔒

Telemetry stays local

Event logs write to ~/.claude/alaude-events.ndjson. UX proposals write to ~/.claude/alaude-ux-proposals.md. Nothing leaves your machine except outbound requests to the providers you picked.

Chat polish

Real markdown, copy buttons on code blocks, hover-to-copy responses, smart auto-scroll, voice input, file attachments (PDF, DOCX, XLSX, images), drag-and-drop. ⌘K to focus, ⌘N for a new session.

🏥

Health Space tools

Lab reference DB, drug interaction checks, PHQ-9 / GAD-7 screening, BMI/BMR calculators, red-flag triage. Informational only — every response carries the clinician-reminder banner.

⚙️

Three-process architecture

Renderer ↔ Electron main ↔ API worker (plain Node with a DNS fallback monkey-patch). Heavy LLM work doesn't block the UI, and VPN DNS resolvers can't break requests to api.anthropic.com.

Seven tuned modes. Plus your own.

Every Space ships with its own system prompt, quick-actions, and (where it makes sense) domain tools. Switch modes in one click. Define custom Spaces in-app for anything the built-ins don't cover.

💬

General

The default. A no-op system prompt — everything else gets domain context layered on top of this.

🩺

Health

Lab result analysis, drug interaction checks, PHQ-9 / GAD-7 screening, BMI/BMR calculators, red-flag triage. Informational only.

💰

Finance

Budgets, invoices, P&L analysis, cash-flow forecasts. Drop in a spreadsheet, ask in plain English.

🏠

Real Estate

Property analysis, MLS listings, investment ROI. Paste a listing URL or PDF, get the numbers that matter.

⚖️

Legal

Contract review, NDA drafting, compliance checks. Plain-language summaries with the sharp edges flagged.

📚

Education

Lesson plans, quizzes, grading, study guides. Scoped to a grade level and framework of your choice.

📣

Marketing

Social posts, email campaigns, SEO, ad copy. Tuned for formats that actually ship.

Custom Spaces

Create your own in-app. Each gets its own system prompt, placeholder copy, and quick-actions.

Models

Local or cloud — the dropdown is the only difference.

When the prompt shouldn't leave your machine, run it on-device. When you need heavier horsepower, route it out. Tool calling is enabled for capable models and disabled for the tiny ones that can't reliably format tool calls.

🦙 Local (Ollama)

  • qwen3.6 · 7B / 14B / 32B
  • gemma-4 · E2B / E4B / 26B / 31B
  • llama-3.2 · 1B / 3B / 11B
  • llama-3.3 · 70B
  • deepseek-r1 · distilled
  • + any Ollama tag

🟠 Anthropic

  • claude-opus-4-7
  • claude-sonnet-4-6
  • claude-haiku-4-5

🟢 OpenAI

  • gpt-4o / gpt-4-turbo
  • o3 · o4-mini

🔵 Google

  • gemini-2.5-pro
  • gemini-2.5-flash

⚫ xAI · 🌙 Moonshot

  • grok-4
  • kimi-k2

☁️ Qwen · 🔷 GLM

  • qwen-max / qwen-plus
  • glm-5.1

Grab the latest build.

v0.2.65 · macOS today · Windows & Linux next.

🍎

macOS · Apple Silicon

M1 / M2 / M3 / M4 · arm64

Download .dmg
🍏

macOS · Intel

x64 · older Macs

Download .dmg
🪟

Windows

coming soon

Setup .exe
🐧

Linux

coming soon

.AppImage · .deb
First launch on macOS. The build is unsigned (no Apple Developer cert yet). Gatekeeper will say "Alaude" cannot be opened because the developer cannot be verified. Either right-click the app → OpenOpen, or run xattr -cr /Applications/Alaude.app once in Terminal. After that, it launches normally.
# or build from source
git clone git@github.com:alsayadi/alaude-desktop.git
cd alaude-desktop
npm install
npm start