Contact
Get in touch.
LinkedIn
GitHub
Email
The deterministic source of truth for your code, infrastructure, and runtime. Query it, assert policies against it, and give your AI agents grounded context instead of guesses.
npx neat.is wires deterministic instrumentation for the frameworks and stores you already run. Anything else can point an OpenTelemetry exporter at NEAT and land in the same graph. Here are a few:
Static extraction reads JavaScript, TypeScript, and Python via tree-sitter. Runtime ingest is plain OTLP (HTTP on :4318, gRPC on :4317), so any service with an OpenTelemetry SDK lands in the same graph.
Every tool you have describes the system after it breaks: logs, dashboards, alerts. None of them hold a model of how the whole thing actually fits together. NEAT keeps that model live, fused from your code and your runtime at once, so your agents work from grounded truth and the answer is there before the incident is.
Most tools start with the change. NEAT starts with the architecture. One command builds a live, deterministic model of your code, your infrastructure, and your runtime: what every service owns, what it calls, what calls it, and what breaks when it fails.
view on GitHubMost systems only see failure after it happens. NEAT makes blast radius a question the architecture answers. get_blast_radius() walks the live edges and returns everything downstream, scored by cascaded confidence. Run it before a deploy, an upgrade, or an agent.
Code says one thing, production does another, and NEAT models both. Every edge is tagged extracted, observed, inferred, or stale. When the two streams disagree, that's a divergence, ranked by impact.
An agent reading raw files is guessing. An agent against a live architecture knows. NEAT exposes the whole model over sixteen MCP tools, from root cause and blast radius to dependencies, divergences, incidents, policies, and semantic search, all refreshed on every save and every span. An agent can check a change against your policies before it writes the code.
see the MCP toolsNEAT builds the model from what you wrote and keeps it honest with what's running. The gap between the two is the finding.
Tree-sitter reads your JavaScript, TypeScript, and Python, pulling out services, calls, schemas, config, and IaC, then lays down the declared graph.
provenance: extractedOpenTelemetry spans stream in over OTLP. NEAT matches live traffic to the graph and marks what it actually sees running.
provenance: observedAsk the graph anything over MCP or HTTP: blast radius, dependencies, divergences, root cause. It refreshes on every save and every span.
provenance: inferred · stale
NEAT runs single-tenant today: npx neat.is on your machine or your own box. Hosted, multi-tenant NEAT is coming. Design partners get it first, and shape what it becomes.