Semantic extraction
Describe the data you want in plain language plus a schema. Pages are located by meaning, not brittle CSS/XPath selectors — extractions keep working when a site’s layout changes.
sortie
Query and act on the web like it were an API. Local-first web agents that drive a real browser and return schema-validated JSON — on your machine, with your choice of LLM.
Multi-step agents
Goal in, browser actions out: navigate, click, type, paginate, recover from failures. Credentials are referenced by name and never shown to the model.
Search, fetch & PDFs
Web search via SearXNG or a browser-engine fallback chain, URL → clean Markdown fetch, and PDF parsing — no LLM key needed for any of them.
Local-first
CLI, typed SDK, REST/WebSocket server, playground UI, and an MCP server. SQLite-backed run queue. No hosted dependency for the core loop — bring Anthropic, OpenAI, or any OpenAI-compatible endpoint (Ollama, vLLM, OpenRouter).