Nudox

Encyclopedia for code

Make AI code generation reliable by grounding agents in compiler-verified, versioned source-of-truth knowledge.

Most efficient way to save tokens during development

By utilizing our graph based database, agents can easily learn and find functions inside of the codebase. Nudox removes the need for expensive grep and search calls that waste thousands of tokens.

How It Works

  1. step 01Scan / Ingest
    We scan public or private repositories (including enterprise private installs).

  2. step 02Compile / Extract
    We derive structured representation of symbols and capture relationships (calls, members, visibility).

  3. step 03Graph + Query
    We store a versioned graph that agents can query via MCP to get compiler-truth facts.

  4. step 04Feedback
    We track what symbols are queried repeatedly and mismatch points to provide analytics.

Core capabilities

Symbol Existence

Know exactly what functions, types, and modules exist in any version of the library. Eliminate hallucinations of non-existent APIs.

Version Awareness

Agents code against specific versions, avoiding hallucinations of deprecated or future APIs. Supports both legacy systems and bleeding-edge libraries.

Lineage & Impact

Understand how symbols change over time and the impact of those changes across the codebase. Enable safer refactors with queryable lineage.

How NuDox compares

Capability comparison between a vanilla LLM, RAG over vector docs, and NuDox.
Capability Vanilla LLM RAG / vector docs NuDox
Version-specific accuracy No Partial Yes
Compiler-verified symbols No No Yes
Hallucination-free APIs No Partial Yes
Lineage & change tracking No No Yes
Private-repo aware No Partial Yes

Drops into your agent

NuDox speaks MCP, so it plugs straight into the tools your team already uses — no new workflow required.

Add NuDox MCP
claude mcp add --transport http nudox https://mcp.nudox.org/mcp \
  --header "NUDOX_API_KEY: YOUR_API_KEY"
// Query the source of truth
const schema = await nudox
  .library("react")
  .version("18.2.0")
  .function("useState")
  .get();
​
// Returns compiler-verified types
schema.params // [initialState: S | (() → S)]
schema.returns // [S, Dispatch<SetStateAction<S>>]// Track lineage across versions
const diff = await symbol.diff("18.1.0");
diff.changed // ["props.variant", "return.type"]
diff.breaking // false
The End of Hallucinations

Eliminate hallucinated APIs and unreliable documentation. Nudox gives AI agents compiler-accurate knowledge of any codebase.

Always Version-Aware

Every query resolves against the exact library version you're running, so agents never guess whether an API still exists in your current release.

Fewer Tokens, Same Model

Agents query the graph directly over MCP instead of burning tokens on repeated grep-and-search — structured facts in, no exploration tax.

Frequently asked questions

What exactly is NuDox?

NuDox is a compiler-backed source-of-truth graph. It gives AI agents and developers version-accurate facts about any library — exact symbol signatures, types, and how they change across releases.

Will Nudox help me save tokens during development?

Absolutely. Nudox is a source-of-truth graph that agents can use to retrieve information. This allows agents to save tokens on expensive grep and search calls.

How is this different from RAG or vector search over docs?

RAG retrieves text that can be outdated, ambiguous, or hallucination-prone. NuDox returns compiler-verified facts tied to an exact version, so answers are precise and checkable rather than "probably right."

Which languages and libraries are supported?

NuDox ingests public and private repositories and compiles them into a versioned graph, with coverage expanding continuously. See how it works above for how ingestion and extraction work.

Does it work with private or enterprise repositories?

Yes. Private installations and hosted private graphs are available on the Enterprise plan — see pricing for details.

How do agents actually query it?

Over MCP. Point your agent at the NuDox MCP endpoint and it can ask for symbol existence, signatures, and version diffs in real time, mid-task.

Is there a free tier?

Public repositories are free to query with standard rate limits. Higher volume and private graphs are covered by the paid plans.