@eonio/agcontext

AGContext (Augmented Context) — a context engineering harness for AI coding agents. Give agents the context a senior engineer would gather before making a change.

AGContext combines code graphs, hybrid retrieval (BM25, embeddings, and graph expansion), repository compression, multi-signal ranking, and token-aware context assembly so any agent — Copilot, Claude Code, Cursor, Cline, or your own — starts each task with the entry points, call sites, contracts, and architecture already in hand.

Works fully offline (deterministic local embeddings, no API keys required); OpenAI / Anthropic / Azure / Google / OpenRouter keys upgrade quality when present.

Install

npm install @eonio/agcontext      # Node.js >= 22

CLI

agc init                                  # scaffold config, gitignore cache
agc index                                 # build/update the index
agc retrieve "how does login work"        # hybrid retrieval + ranking
agc context "how does login work" --format xml --budget 8000
agc explain AuthService                   # relations, metrics, summary
agc doctor                                # health checks

Library

import { AGContext } from "@eonio/agcontext";

const agc = new AGContext();
await agc.index();

const result = await agc.retrieve({ query: "How does authentication work?" });
const pkg = await agc.context({ query: "How does authentication work?", maxTokens: 8000 });
const xml = await agc.contextText({ query: "…", format: "xml" });

Configuration lives in agcontext.config.ts:

import { defineConfig } from "@eonio/agcontext";

export default defineConfig({
  graphDepth: 3,
  maxNodes: 50,
  ranking: "hybrid",
});

Documentation

Full docs — product vision, architecture, API reference, CLI guide, configuration, providers, plugins, performance, roadmap — live in the GitHub repository.

License

MIT © eonio