Skip to main content
Contexity is a local-first context continuity layer for AI coding agents. It stores structured project knowledge — decisions, investigations, external requirements — and delivers bounded, task-aware context packs so agents like Codex, Claude Code, and Cursor always work with the right information at the right time.

Quick Start

Attach Contexity to a project and connect your first agent host in minutes.

Install

Build Contexity from source and configure your local state home.

How It Works

Understand the core loop: attach, retrieve, work, and close.

Trust & Freshness

Learn how Contexity prevents stale or unverified context from steering agents.

Use with Codex

Connect Contexity to Codex through MCP for automatic context retrieval.

Use with Claude Code

Set up Claude Code to call Contexity tools automatically during tasks.

CLI Reference

Every Contexity command for setup, inspection, and automation.

MCP Tools

Agent-facing tools exposed through the Contexity MCP server.

​
The Core Loop

Contexity follows a four-step loop on every agent task:
1

Attach

Contexity creates a project identity and local state for your repository. Run this once per project checkout.
2

Retrieve

Before doing work, the agent asks Contexity for a compact, freshness-aware context pack scoped to the current task.
3

Work

The agent edits, analyzes, and tests with the retrieved context. Contexity maps changed files through the project graph as work progresses.
4

Close

At the end of a task, the agent records what was useful, what changed, any unresolved threads, and heuristic metrics.
Contexity is local-first. All project state lives in .cxcap/ and your local Contexity state home — no cloud account required.