Archaea Biblio · The knowledge layer
Make AI know what your company knows
Biblio is an enterprise-grade retrieval augmentation system for AI agents. It turns scattered documents, database tables, code, and third-party system data into clean, precise context your AI can answer from — and none of it leaves your perimeter.
HEADLESS API + WEB CONSOLE · SELF-HOSTED · KNOWLEDGE GRAPH INSIDE
The problem
Your GPT has a RAG layer slapped on it — and the answers are off
Generic retrieval fails on real enterprise material: hundred-page PDFs, dense tables, legacy code, data scattered across systems. Biblio is built for exactly that material, and it runs where compliance requires — on your own servers.
The flow-lite document engine
Archaea's own parsing engine handles PDF, Office files, and images with OCR, deep document decomposition, table extraction, and AI semantic understanding. It's the same engine embedded inside every ArcFact deployment — one document intelligence core across the whole stack.
Six retrieval modes, one corpus
Local, global, mix, graph, vector, and bypass modes cover the full spectrum — from pinpoint fact lookup to cross-corpus reasoning over the knowledge graph. Your agents get the right retrieval strategy for each question, not a one-size-fits-all similarity search.
Vectors plus entities and relationships
Biblio builds a knowledge graph alongside the vector store, connecting entities across documents, schemas, and code. Answers draw on structure, not just similarity — which is what lifts retrieval accuracy on complex, interlinked enterprise material.
Seven built-in connectors, plugin-extensible
Git, Google Drive, Oracle, PostgreSQL, S3, direct upload, and web crawl ship in the box. Data is pulled in and pre-indexed for offline, high-speed retrieval; new source types are added as plugins.
Headless API, built to be called
Biblio is a backend your developers integrate — REST API and SDK-style access, plus a web console for knowledge librarians. It serves ArcFact workers over MCP, and it serves any third-party agent framework the same way.
Bring your own model
Biblio binds to the LLM you choose — a fine-tuned model, open-source models on your own hardware, or a MaaS endpoint. No model lock-in, and no requirement to move data to reach a model.
Deployment
Light to run, serious about isolation
| Deployment | Single-host docker-compose; one to two mid-size servers to start |
| Scaling | Horizontal — add distributed task workers as volume grows |
| Storage | PostgreSQL vectors · Neo4j knowledge graph · Redis cache |
| Access control | JWT plus LDAP; multi-tenant isolation at the Collection level |
| Ops burden | Low to moderate — conventional database operations |
Who runs Biblio
Biblio is an infrastructure decision, typically owned by the Chief Data Officer, the head of AI platform, or the knowledge management lead. The evaluation is technical and fast: point it at a slice of your corpus and measure retrieval accuracy against what you have today.
When you need it
Biblio pays off when your corpus is measured in the tens of thousands of documents and your strategy calls for an AI and data backbone you own — not a public RAG SaaS your compliance team can't approve.
Save the engineer-years of building RAG yourself, lift retrieval accuracy on the documents that matter, and keep every byte inside your perimeter. The Biblio case, in one sentence
Start with a fast, low-risk POC
Bring a slice of your corpus. We'll show you what precise retrieval looks like on your own material — then talk about what workers could do with it.