Everyone has the same models. Your edge is context.
Oxagen turns your business data and your code into one connected map. Your read it to get the right answer the first time — so you can run them on smaller, cheaper models.
npx @oxagen/mcp-install init --agent cursor-native · -backed · -scoped · included
95%
Inference cost reduction
<50ms
p95 3-hop graph retrieval
2 graphs
Business ontology + codebase
Evals
Built in, not bolted on
Context is the moat
Every team has the same models. Not the same context.
GPT-4o, Claude, Gemini — anyone can use them. The teams that win aren't paying more per question. They're feeding their agents better context so they can pay less. A model with no memory has to guess what Order means in your system and which service owns it. An agent reading your Oxagen graph knows. Swap for , cut 95% of , and get a more accurate answer.
Business ontology
Entities, relationships, and domain schema auto-discovered from your data sources. Agents traverse your business model, not a flat vector index.
Code graph
Classes, functions, data models, and call graphs. Agents understand your implementation — not just your docs — so they stop hallucinating APIs that don't exist.
Built-in evals
Compare accuracy and cost per query across model tiers. See the delta. The graph is the argument; the evals are the proof your CTO needs to approve the infrastructure.
How it works
Graph. Query. Downgrade your model.
Step 1 · Connect
Plug in your data and your code
Point Oxagen at the systems you already use — CRM, docs, GitHub, finance, the works. Oxagen reads them and builds one connected map of your business and your codebase.
Step 2 · Read
Agents read the map
Your agents look up the map over MCP, the same way Cursor or Claude Desktop plugs into a tool. They know your business and your code instead of guessing.
Step 3 · Save
Run on cheaper models. Prove it works.
Better context lets you swap the biggest model for a smaller one. We measure accuracy and cost before and after — so you can show the savings, not just claim them.
Context layer
The map your agents should have been reading all along.
Your code has structure. Your business has rules. Your agents should know both — not rediscover them on every question. Oxagen turns both into one connected map your agents can walk.
- Business map — your customers, accounts, orders, and how they connect, picked up automatically from the systems you already use
- Code map — every class, function, and call path in your repos, so agents know what your code actually does
- Fast lookups across both — under 50ms, even when the answer takes three or four hops to find
Connections
Plug it in once. Keep reading forever.
Connect a system once and Oxagen keeps the map up to date as your data changes. Every agent you build later gets a live view — you don't have to babysit a pipeline.
- One-click connectors for email, calendar, finance, storage, and code repos
- Other tools and agents can also feed in over MCP
- Add a new source class without writing custom glue code
Agents
One shared memory for every agent.
Stop giving each agent a blank slate. Every agent reads the same map — same customers, same code, same history. They all agree on what an Account is.
- Claude, ChatGPT, and your own stack all plug in over the same MCP server
- Every agent refers to the same record across runs — no duplicates, no confusion
- Stream reasoning back in real time, or ask precise questions with a typed SDK
Evaluations
See the savings. Prove the accuracy.
Evals come built in, not bolted on. Run the same agent with and without the Oxagen map. Watch accuracy go up and cost go down — on the same dashboard.
- Side-by-side scores: same question, with the map vs. without
- Cost per question tracked across model sizes — show finance the 95% drop
- Re-runs every time the map updates, so accuracy gets better over time
Security
Built for many customers, separated by design.
Each customer's data lives behind hard walls the database itself enforces. No agent can read across the line. Your data never trains anyone's model.
- Row-level security — the database blocks queries that cross customer lines
- OAuth tokens and graph data encrypted at rest (AES-256-GCM); TLS 1.3 in transit
- Your data is never used to train models. SOC 2, GDPR, and HIPAA work in progress.
From the blog
Engineering notes on agent infrastructure.
Run faster models. Get better results.
Graph your context. Let agents traverse it. Ship evals that prove the delta — in accuracy, in cost, in agent performance.
No sales call required. Self-serve from install to first query.