Anchor is an internal AI support and research agent that processes documents and answers user queries, escalating to humans when necessary. It is a multi-tenant SaaS application built with Python, TypeScript, HTML, CSS, and Dockerfile, featuring PDF ingestion, local embeddings, and automated evaluation.
From the README
Anchor An internal AI support and research agent that answers questions from your own documents — and escalates to a human when it can't. Anchor is a production-shaped RAG system: it ingests PDFs (including scans), embeds them locally, retrieves relevant passages, routes each question to an appropriate LLM provider, lets the model call internal tools, validates the answer on the way out, and records everything it does in structured logs and metrics. An automated evaluation suite catches quality regressions. It is a multi-tenant SaaS application: users and workspaces, per-workspace documents, conversations, API keys, analytics and audit trails, with Argon2id passwords, rotating refresh...
Skills
- Python
- TypeScript
- HTML
- CSS
- Dockerfile