Introducing Fluree AI: One Place for Your Data and Context — Infinite Interfaces on Top

Somebody on your team already did this. They took a spreadsheet — the customer list, last quarter’s contracts — and dragged it into a chatbot. It answered questions. It felt like magic. It felt like we’re doing AI on our data now.

Then tomorrow came. Someone asked for a follow-up, and the chatbot answered from the stale file it still had open. A colleague who should never have seen the compensation column opened the same upload and saw everything — because a file doesn’t know who’s asking.

Uploading a file to a chatbot isn’t a data strategy. It’s a demo that works once.

Today we’re launching the thing that works every day after: Fluree AI is now generally available. Connect your data sources, and it automatically builds a governed knowledge graph — an institutional memory for your organization that every person, application, and AI agent can query and trust. Natural language in, verified and cited answers out. No infrastructure to stand up, no cluster to size, nothing to patch at 2 a.m.

It runs on FlureeDB — our secure, trusted, and verifiable knowledge graph database.

Connect what you actually have

Fluree AI meets your data where it lives:

From a folder of files to a governed graph — automatically

Point Fluree AI at your data and it does the work no one wants to do by hand. It profiles every column — types, uniqueness, null ratios. It finds the seams between your files: which ID in one table is really a foreign key into another. It scores relationships across all your sources at once and drafts a real ontology — classes, properties, identity strategies — validated for consistency before a human ever sees it.

Then it stops. Nothing reaches your live graph without your sign-off. The system surfaces exactly the judgment calls it’s least confident about — the ones with the highest blast radius — and puts them in front of you. The tedious 90% happens without you; the 10% that’s actually a decision stays yours.

That’s what “automatic” should mean. Automatic isn’t the same as unsupervised.

One record per customer, finally

Every company has the same mess: “IBM,” “International Business Machines,” and “I.B.M. Corp” living as three records in three systems. Fluree AI runs true probabilistic record linkage — the Fellegi-Sunter model, the statistical gold standard — across people, companies, and products. Every match comes with the evidence behind it, not just an unexplained similarity score. Confident matches merge on their own; ambiguous ones go to a human review queue. And once a person makes a call, no future model update quietly reverses it.

Golden records, without the six-month MDM project.

Ask it however you work

One graph, every interface:

• Natural language for anyone. Ask a question, get a verified answer with citations to the actual source — not a paraphrase, and never data the asker isn’t entitled to see.
• Live dashboards in plain English. Ask for a dashboard and get compiled, live code that re-queries the graph every time it opens — and can be rewound to any past commit. “What did we know the day we made that decision?” is a click, not a forensic project.
• A TypeScript SDK and REST API for builders. Fork a starter template, open it in Claude Code or Cursor, and ask for the app you want. Every app runs on scoped, self-refreshing tokens — even a buggy app physically cannot reach data it wasn’t granted.

Built for agents, not just for chat

Fluree AI is native to the Model Context Protocol. Claude, Cursor, and any MCP client connect to your graph as a first-class tool — setup is copy, paste, sign in. No custom integration, no shared secret sitting around to leak: every agent connects as an authenticated user and sees only the intersection of what the workspace exposes and what that person is allowed to see.

And there’s a mutation gate. Write-capable tools are withheld from an agent entirely until a human clears them, and approval is checked mid-run — not just at connection time. Your agents get institutional memory; you keep the keys.

This is the difference between an agent that’s plausible and an agent that’s aware — the Context-in-the-Loop architecture we’ve been writing about, now available as a product. Because agents finally share one governed graph instead of a pile of private context windows — the same agentic memory problem that makes long agent sessions expensive also makes them ungrounded — cross-functional work like M&A due diligence, enterprise risk assessment, and regulatory compliance becomes automatable for the first time.

Trust is built in, not bolted on

Everything Fluree AI does inherits FlureeDB’s trust layer:

• Always current, never a snapshot. Answers come from what’s true in the graph right now, not the file someone uploaded in March.
• Per-user policy, enforced in the data. Security lives at the most granular level of the graph itself, so permissions follow the data into every query, every dashboard, every agent conversation. Two users asking the same question get appropriately different answers — automatically.
• Every answer carries its receipts. Full provenance and history: who asserted a fact, when, and from what source. Every dataset is time-travelable to any point in its past.

This is what “AI-safe data” means in practice — and it’s why grounding LLMs in a knowledge graph instead of a data silo doesn’t just make them more accurate, it makes them dramatically more token-efficient.

Enterprise-grade from day one

Fluree AI is fully serverless and scales horizontally with demand — a dormant graph costs next to nothing, and importing a large dataset never blocks anyone’s queries. Each customer’s data lives in a genuinely isolated environment, not a shared schema with row filters standing in for real isolation. Enterprises with strict requirements can deploy the entire stack inside their own AWS account as a versioned, self-contained deployment.

It’s the same technology already powering mission-critical workloads at the U.S. Department of Defense, Morgan Stanley, The Associated Press, Dow Jones, Warner Bros. Discovery, and Arizona State University. Fluree was named a Gartner Cool Vendor in AI Data Management.

Your first answer in minutes

Signing up provisions a real, dedicated environment automatically — no sales call, no queue, no credit card. Connect a source, ask a question, build a dashboard. The documented path from sign-up to first insight is about 15 minutes; your first query runs in under 30 seconds.

The difference between a demo and a data strategy is what happens on day two. Start day two today.

Prefer to run the engine yourself? FlureeDB is open source. Coming from Neo4j, Jena, Stardog, or GraphDB? There’s a migration guide for each. Enterprise isolation requirements? Talk to us.