Context Architecture for AI-Assisted Development

Deterministic context assembly for AI coding agents. No RAG, no embeddings — just structured knowledge that follows explicit links.

Get Started View on GitHub

Why Contextia?

AI coding agents are powerful, but they’re blind without context. They don’t know which spec governs a file, why a decision was made, or what norms apply. Contextia solves this with deterministic context assembly — the agent asks for context, and gets exactly the right documents, every time.

Deterministic, not probabilistic

No embeddings, no vector search, no hallucinated results. Context follows explicit links: task → spec → decisions → norms.

Two interfaces, one core

CLI for humans (bootstrap, CI/CD, maintenance). MCP server for AI agents (direct tool access via Model Context Protocol).

Markdown all the way

Every artifact is Markdown + YAML frontmatter. Human-readable, version-controllable, editable without tooling.

Language-aware scanning

Tree-sitter powered annotation scanner detects @spec in comments, identifies enclosing classes and functions.

How it works

Your project
├── .contextia/
│   ├── system/          # Durable knowledge (specs, decisions, norms)
│   ├── work/            # Operational state (tasks, plans, logs)
│   └── config.yaml      # Project configuration
└── src/
    └── auth.py          # @spec SPEC-AUTH-001 ← bidirectional link

The agent calls contextia context TASK-001 and gets back exactly the specs, decisions, and norms needed for that task — assembled deterministically by following explicit links, not by guessing.

Introduction Learn what Contextia is and why it exists

Quick Start Set up your first project in 5 minutes

CLI Reference All commands documented

MCP Server Connect your AI agent