Under the Hood

How it works

From raw repository source code to an honest, evidence-backed knowledge layer for AI coding agents.

Analyze

Scan the repo locally

Athena visits project files, package configurations, AST routes, database schemas, and git metadata entirely on your machine. Secrets are detected and immediately redacted.

Inspects package manifests, tsconfig, Dockerfiles, and CI pipelines
Traverses HTTP routing trees and database entities (e.g. Prisma, SQL)
Analysis, documentation, the web UI and MCP server run fully offline

athena init

Write

Write .athena/*.md knowledge with evidence

Athena creates structured markdown files in .athena/. Every fact is labeled FACT, DETECTED, INFERRED, or UNKNOWN with file paths (and line numbers where available).

11 generated documents (architecture, database, API, auth and more) plus your own rules.md
Preserves developer notes and custom guidelines across re-analysis
Refuses to guess: missing details are labeled UNKNOWN

athena analyze

Connect

Agent instructions, hooks, and MCP server

Integrate with Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI, Antigravity, Windsurf, Cline, or any agent using AGENTS.md. Agents start Athena's MCP server (`athena mcp`, 12 tools) whenever they need context.

Hooks log agent activity so you can inspect tool usage locally
MCP server enables agents to fetch specific schemas or rules on demand
Standard AGENTS.md file provides a universal anchor for all models

athena agents add <id>

Sync

Propose updates as code changes, asks once before applying

athena sync shows which documents would change and why, then asks once before applying them. Sections you edited by hand are kept unless you pass --force.

athena watch monitors files in the background and proposes updates
Nothing is ever silently overwritten
Supports --dry-run, --diff, --yes, and --check for CI automation

athena sync

Local RepoAthena AST Analyzer.athena/*.md KnowledgeClaude Code · Cursor · Codex · Copilot · Gemini CLI · and more

Interactive Data Pipeline

Best-in-class context.

See how Athena turns raw repository files into evidence-backed, honest intelligence that AI coding agents rely on instead of guessing.

Active Stream:ORM & Entities

Phase: input·100% Localhost Execution

01. Repository Source

Prisma AST Entityprisma/schema.prisma:1

model User { id String @id email String @unique }

02. Athena Engine

AST Entity Parser

Resolved models, fields & relations without guessing

Assigned Label:DETECTED

03. Agent Consumption

.athena/database.md

MCP Tool

Query tool: get_database_schema

✓ User model [DETECTED: prisma/schema.prisma:1] -> 2 fields

Delivered to:Cursor IDE & Claude Code

Verifiable Receipt Citation: prisma/schema.prisma:1 → .athena/database.md

Command:athena mcp

Guarantees & Boundaries

What Athena never does

Athena is built on the principle of transparency. These are our non-negotiable boundaries.

Never sends code anywhere for core analysis

Analysis, documentation, the web UI and the MCP server run fully offline. Two optional commands use the network, and only when you run them: athena security runs your ecosystem's audit tools (such as npm audit), which query their own advisory databases, and athena ai enrich sends redacted documents to the AI provider you configured.

Never stores secret values

Athena flags environment variable names for context, but automatically redacts and omits secret values.

Never overwrites your edits

Re-running athena analyze or athena sync preserves any manual notes, rules, or edits you made in .athena/ files, unless you explicitly pass --force.

Never claims something it can't show evidence for

Every statement carries a FACT, DETECTED or INFERRED label with its evidence. If evidence is missing, it explicitly marks UNKNOWN.

100% open source under the MIT License on GitHub

Ready to give your AI agent real codebase memory?

Install project-athena globally or run it directly with npx.

Quick start guide View all features