$ curl -fsSL https://spaceduck.ai/install.sh | bash
Works on macOS, Linux, and Windows. See the full quickstart guide for provider setup and optional tools.
Hybrid recall (vector + keyword) finds what you said even when you don't use the same words. Facts are extracted eagerly after every response.
Swap between local models (llama.cpp, LM Studio) and cloud providers (Bedrock, Gemini, OpenRouter) from the Settings UI — no restart required.
Web search, browser automation, document scanning, and HTTP fetch. The agent loop chains tool calls automatically.
Web UI, Desktop app (Tauri), WhatsApp, and CLI. Same memory, same tools, any surface.
Schedule tasks on cron or interval. The agent runs in the background, routes results to memory or notifications, retries on failure.
Per-task token limits, cost caps, and wall-clock timeouts. Global daily and monthly budgets pause the scheduler before you overspend. Not a plugin.
Automatic fact extraction with correction semantics. Memory is inspectable, correctable, and yours.
Switch between local and cloud models at runtime. Chat and embeddings are independent subsystems.
Brave, Perplexity, or SearXNG. The agent decides when to search and weaves results into the conversation.
Playwright-powered browsing. Navigate, click, extract data, and interact with any website.
Upload PDFs via the chat UI. Marker converts them to markdown so the assistant can read and discuss them.
Whisper-powered speech-to-text. Hold the mic button to record, release to transcribe. Runs locally.
Cron jobs, intervals, or one-off runs. Priority queue with dead-letter handling. View status, cost, and results in the built-in dashboard.
Drop a SKILL.md file and it just works. Security scanner blocks injection attempts. Tool scoping limits what each skill can access.