Uvi: the context layer for enterprise AI
#Introducing Uvi
Today, we're introducing Uvi. The only context layer that sits underneath your agents and maps the relationships between your systems, entities and decisions, so they stop guessing at your context. (Some of you knew us as Naboo, same team, same product, new name).
We started in 2024 building infrastructure for a problem most companies don't have a name for yet. Since then we've deployed with enterprise engineering teams, and recently closed a $7.5M seed round to expand it.
#Why now?
Results haven't kept pace with how fast businesses want to use AI agents.
AI is constantly playing catch-up, guessing at why something is the way it is because the reasoning behind it was never captured, re-figuring out the same relationships every other agent in the company already worked out on its own, and surfacing answers that are stale or wrong with no signal that they are.
So an agent burns more time and more tokens getting there, and still lands on a response that's wrong.
The answer isn't more context. It's what the company already decided, held as a chain of decisions, owners, triggers and blockers that produced the current state. Some people call this lineage: the state you see is the result of steps, and the steps are the part nobody wrote down.
That's why we built Uvi:
The infrastructure layer that keeps that chain living, shared across the enterprise, that any agent can query directly.
#How it works
Uvi connects to the systems your business already runs on, your codebase, PRs, tickets, docs, agent sessions, and conversations, to build a living graph from them. Not a copy of your documents, a map of the decisions inside them: what was decided, who owns it, what triggered it, and what's blocking it. Getting there means clearing five walls: resolving entities across systems, learning your vocabulary, understanding intent, enforcing real permissions, and keeping all of it current under load.
Agents query that graph directly instead of searching your systems and inferring the rest. The relationships are already resolved, so an agent that used to take dozens of round trips to piece something together can get there in a few.
Uvi inherits the access controls you already have. It doesn't introduce a second permission system to maintain, and it doesn't ask you to re-map your org into a new one. Permissions are enforced at query time at the graph level, so an agent only ever returns what the person behind it can already see. Every query leaves a trace, so you can audit what an agent asked for, what it got back, and what it cost.
It runs as SaaS or on-prem.
#What makes Uvi different?
Decisions, with the reasoning attached: Most tools retrieve what was written down. Uvi holds what was decided, with the owner, the trigger, and the blockers attached, so an agent can follow the reasoning instead of inferring it from whatever file it happened to find.
Resolved before the question, narrowed by intent: The relationships are worked out ahead of time. An agent isn't running a live investigation across your systems every time someone asks something, which is where most of the cost and latency comes from.
Infrastructure that's not a black box: You define your own entities and keep visibility into traces and cost. One customer's technical team told us plug-and-play was generous, but what they actually wanted was control. That's the version we built.
#Who is Uvi for?
Uvi is built for companies where:
- Answers live across systems, never in one. Not just repos, tickets and threads, but the entities your business actually runs on: services, clients, KPIs, agent sessions.
- The questions that matter need the reasoning behind a decision, not just the document that followed it
- Someone already tried building this internally, and it solved access without solving relevance
- Security and permissions actually matter, so an agent can't surface something the person asking shouldn't see
#What you can expect
We tested Uvi against the alternative directly: more than 900 real employee questions, run through the same frontier model two ways. Agents built the way most enterprises build them cost roughly 8x as much per answer and took more than twice as long. On open-ended questions, where agents got stuck in search loops, it was 46x.
#Get started
If your agents are working across your systems and still guessing, we'd like to show you the results you actually deployed them for.