Civic AI - decades of records, answered with the citation attached
The records officer knows the answer exists: in committee minutes filed thirty years ago, in a policy draft that was superseded twice, somewhere across the systems the archive has passed through. Civic AI, built on Blue Mesh's Enterprise RAG, indexes those archives where they live and answers questions asked in plain language, with every answer citing the document it came from.
One archive, three ways to interrogate it
A public archive is not a search problem the way a website is. The documents are long, the language is procedural, and the cost of a confident wrong answer is a decision built on it. Every capability here is built around the citation for that reason.
Natural-language search across the archive
Retrieval-augmented search across policy drafts, legislative records and discussions, indexed where they already live. Answers cite their source.
- 01.01Plain-language questions over procedural documents
- 01.02Archives indexed in place
- 01.03Answers grounded in retrieved passages
- 01.04Source cited on every answer
Policy comparison
The engine reads two texts against each other: a draft against the policy it replaces, a bill against the act it amends. Divergences cite both passages.
- 02.01Draft against predecessor, side by side
- 02.02Divergences surfaced, not summarized away
- 02.03Both passages cited on every finding
- 02.04Reads the texts the analyst would have to read
Conversational Q&A, on the record
Follow-up questions keep their context, so a line of inquiry runs as a conversation rather than cold searches. Every query and answer is logged.
- 03.01Context held across follow-up questions
- 03.02Accessible on any platform
- 03.03Every query and answer logged
- 03.04A trail a records office can stand behind
The archive that answers questions
Decades of drafts and records across many systems. Finding anything was slow, and where things sit retires with the people who know.
Blue Mesh introduced Civic AI, on the Privy stack: RAG search, a policy comparison engine and conversational Q&A across large document sets.
What began with knowing which cabinet to open now starts with a question and ends at a cited passage, and answers a new hire the way it answers a veteran.
Built for institutions that answer to the public
Answers that carry their source
A public body cannot act on "the system said so." Every answer cites the passage it stands on, so the officer verifies against the original before the answer informs advice, analysis, or a decision. The citation is not a nicety; it is the mechanism that makes the answer usable.
The record of the record
Every query and answer is logged for compliance audit, attributable and timestamped. When someone asks what informed a piece of analysis, the trail exists: what was asked, what was retrieved, what came back.
Deployment a public archive can accept
The platform runs private cloud, on-premises, air-gapped. For archives that cannot sit on a connected network at all, BM in a Box ships it pre-configured on NVIDIA DGX hardware with no path to the internet.
Where it stops
Civic AI answers from what the archive contains. It cannot recover what was never written down, and a misfiled document produces a well-cited wrong answer, which is exactly why the citation is there. Verification stays with a person, and the decision stays with the officeholder.
- Privy AI The Enterprise RAG engine Civic AI is built on: retrieval from your own indexed content, grounded and citation-backed.
- BM in a Box For archives that cannot sit on a connected network: the platform pre-configured on NVIDIA DGX hardware, running air-gapped.
- Trust and security The page a public body puts in front of its own reviewers: deployment modes, named controls, an audit trail on every decision, and where certification honestly stands.
Bring the question your archive cannot answer quickly
Point Civic AI at a real corpus: minutes, drafts, records. Ask it live, read the citations for yourself, and see the log every question leaves behind.