Centy products / 02
Your docs. Your server. Your agents.
Give an AI agent a narrow, read-only way to search your documents, answer with evidence, and notice when sources change.
Python 3.10+ · Ubuntu-ready · no public port required
What changed in the latest policy?
The updated policy changes the approval step for reimbursements [1].
Page 4 · indexed snapshot
Illustrative output — agents should verify consequential answers.
Built for a smaller trust boundary
A library card for your agent.
The agent gets answers and references, not a tool that can rewrite or upload your files. You decide which documents sit in its dedicated collection.
Add your sources
Put PDFs, Markdown, or text files in a private folder. Index them from the admin command line.
Connect an agent
Run the MCP server locally or over SSH. It exposes search, cited answers, and source status only.
Keep the boundary
No public web endpoint is needed. Use a separate Unix account and document folder for each collection.
From DocChat, with a new job
Already have a DocChat index?
Import an index you created with DocChat to reuse its Voyage embeddings without paying to index again. A fresh ingest adds PDF page references and change tracking.
The index stays on your server, but ingestion sends document chunks to Voyage AI. Searches send queries to Voyage AI; generated answers send retrieved excerpts to Anthropic. Review those data flows before using sensitive files.
Meet DocChatReady to try it?
Start on your own server.
The download includes Ubuntu setup, an SSH connection example, and synthetic tests. Bring your own API keys. No documents are included.