Blog
Local-first data, governed agent access, data contracts, and DuckDB federation, from the maker of DataCharter.
August 12, 2026
An AI agent reading your data is two trust problems at once: can you prove its answer, and can it trust the data? DataCharter now does both — signed, verifiable answer receipts and prompt-injection quarantine at the data plane.
Read it →August 7, 2026
Seven features in a week meant new panels that opened empty for first-time visitors. DataCharter 0.18.0 ships a demo workspace where everything is real — and an 11-step tour that walks you through it.
Read it →August 5, 2026
Google's clean room won't return a row unless 50 users stand behind it. DataCharter 0.17.0 brings that class of control home: plain-English policies — aggregates only, groups of at least k, no joins to X — enforced by query analysis.
Read it →August 3, 2026
You already wrote the tribal knowledge — one WHERE clause at a time. DataCharter now mines your query history for the habits you repeat and turns them into agent guides, with evidence, fully offline.
Read it →August 2, 2026
The governed data explorer is now a double-clickable app — macOS and Windows, beta, no terminal required. Same local server, same masking, same audit chain, now with a dock icon.
Read it →August 1, 2026
Security folks have loved honeytokens for decades: any touch is by definition suspicious. DataCharter 0.15.0 plants them for AI agents — masked canaries that can only surface if your protection layer actually failed.
Read it →August 1, 2026
Most teams running AI agents can't answer 'what did it touch?'. DataCharter 0.14.0 ships the missing piece: a tamper-evident audit trail of every agent query — dual-attributed, hash-chained, exportable as evidence.
Read it →July 29, 2026
A week ago I argued that agent context belongs in the contract. Fair question back: how do you know it helps? DataCharter 0.12.0 answers it — eval suites that score the agent on your data and show exactly how much your guides moved the needle.
Read it →July 29, 2026
MotherDuck just showed that prose context is the biggest lever for agent SQL accuracy. Here's the local, contract-governed take: guides that live in your repo and ship through the governed surface — in DataCharter 0.11.0, out today.
Read it →July 28, 2026
Handing an LLM your database credentials is indefensible. Pasting CSVs into a chat is a joke. Here's the third option: a data contract the agent physically cannot violate.
Read it →