CxMS

<div align="center">

<h1>🚀 CxMS Pro is Here</h1>
<h3>The AI That Never Forgets</h3>

**Your corrections stick. Your decisions persist. Every session builds on the last.**

Persistent memory with confidence-based knowledge tiers, accuracy enforcement, and self-improving recall. Born from 184+ real production sessions. 16 patents pending. Zero cloud dependency.

$49/mo per seat (1 AI assistant included, +$19/mo per additional agent)

Works with Claude Code, Cursor, Windsurf, GitHub Copilot, Cline, Continue.dev, Zed, and Amazon Q.

[![Get CxMS Pro](https://img.shields.io/badge/Get_CxMS_Pro-$49/mo-6366f1?style=for-the-badge)](https://opencxms.org/products/cxms-pro) [![Try Core Free](https://img.shields.io/badge/CxMS_Core-Free_&_Open_Source-22c55e?style=for-the-badge)](https://github.com/RobSB2/CxMS)

| | CxMS Core (Free) | CxMS Pro |
|---|---|---|
| Full-text search | :white_check_mark: | :white_check_mark: |
| Session lifecycle hooks | :white_check_mark: | :white_check_mark: |
| Context monitoring | :white_check_mark: | :white_check_mark: |
| **Persistent memory engine** | | :star: |
| **Confidence-based knowledge tiers** | | :star: |
| **Accuracy enforcement** | | :star: |
| **Automatic context awareness** | | :star: |
| **Source verification before writes** | | :star: |
| **Self-improving memory** | | :star: |
| **Conflict detection and resolution** | | :star: |
| **Cross-project knowledge sharing** | | :star: |
| **Adaptive context scaling (128K-1M)** | | :star: |
| **Project onboarding + migration** | | :star: |
| **9 AI assistant support** | | :star: |

</div>

---

# CxMS - OpenCxMS Memory System

<div align="center">

[![Website](https://img.shields.io/badge/Website-opencxms.org-blue?style=for-the-badge)](https://opencxms.org)
[![Version](https://img.shields.io/badge/Version-3.0.0-purple?style=for-the-badge)](templates/VERSIONS.md)
[![Contact](https://img.shields.io/badge/Contact-opencxms%40proton.me-orange?style=for-the-badge)](mailto:opencxms@proton.me)

**Persistent memory for AI coding assistants through structured documentation.**

*Because AI context is temporary, but files are permanent.*

</div>

> **Note:** This repository was restructured and relaunched in early 2026. Current GitHub analytics (clone count, stars, traffic) reflect only the period since the relaunch and do not capture the full adoption history across prior versions. Total clones since inception exceed 1,000.

---

## The Problem

Every time you start a new session with an AI coding assistant:

> "Continue working on the authentication feature we discussed Friday."

> "I don't have any context about previous conversations. Could you tell me about the authentication feature?"

You spend 15-30 minutes re-explaining your project, decisions, and progress. **Every. Single. Time.**

```
WITHOUT CxMS                              WITH CxMS

Session 1  Session 2  Session 3          Session 1  Session 2  Session 3
    |          |          |                  |          |          |
    v          v          v                  v          v          v
+-------+  +-------+  +-------+          +-------+  +-------+  +-------+
| FULL  |  | FULL  |  | FULL  |          | FULL  |  | FULL  |  | FULL  |
|CONTEXT|  |CONTEXT|  |CONTEXT|          |CONTEXT|  |CONTEXT|  |CONTEXT|
|       |  |       |  |       |          |   ^   |  |   ^   |  |   ^   |
|  ^    |  |  ^    |  |  ^    |          |   |   |  |   |   |  |   |   |
|  |    |  |  |    |  |  |    |          | 5 sec |  | 5 sec |  | 5 sec |
|15-30  |  |15-30  |  |15-30  |          | read  |  | read  |  | read  |
| min   |  | min   |  | min   |          +---+---+  +---+---+  +---+---+
|rebuild|  |rebuild|  |rebuild|              |          |          |
+-------+  +-------+  +-------+              v          v          v
                                         +-----------------------------+
    ====================                 |    Session.md persists      |
    ~60-90 min WASTED                    |    context between sessions |
    per week                             +-----------------------------+
                                             ~0 min wasted
```

## The Solution

**CxMS gives AI assistants structured, persistent, user-controlled memory through markdown files.**

| Without CxMS | With CxMS |
|--------------|-----------|
| Start from zero each time | Pick up exactly where you left off |
| Re-explain everything | AI reads context files automatically |
| Decisions forgotten | Decisions documented with rationale |
| Inconsistent approaches | Consistent patterns maintained |

---

## What's New

### v3.0.0 — Total Recall v2: Session ID Coordination (February 19, 2026)

Multiple Claude Code sessions on the same project are now tracked individually. The coordination file gains an `active_sessions` array (schema 1.1, auto-migrated) that records each session's ID, status, context %, and model.

- **Concurrent session detection:** Starting a second session on the same project triggers a `CONCURRENT SESSIONS DETECTED` banner with the sibling's age and context usage
- **Session lifecycle:** SessionEnd marks sessions `completed` with final stats. Stale sessions auto-pruned (completed >1hr, active >4hr)
- **Per-session message tracking:** Coordination messages record which specific session read them (`{project, session_id, read_at}` objects), backward-compatible with existing string format
- **Per-session cleanup:** State files deleted on session exit; SessionStart only clears its own session's files (no longer destroys concurrent sessions' state)

Updated: `cxms-session-start.mjs` (v3.1 → v4.0), `cxms-session-end.mjs` (v2.1 → v2.2)

### v2.0.5 — Per-Session Context Isolation (February 19, 2026)

Fixed a bug where two sessions on the same project shared a single `context-status.json`. Each session now writes to its own `context-status-{session_id}.json`. Compaction detection logging added.

### v2.0.0 — Enforced Startup + Total Recall (February 8, 2026)

Major release: PreToolUse hook blocks all non-Read tools until required startup files are confirmed read. Cross-instance memory transfer via coordination file. Full changelog at [opencxms.org/updates](https://opencxms.org/updates).

---

### Works With Any AI Assistant

CxMS is **AI-agnostic**. It works with any coding assistant that can read files:

| AI Assistant | Compatible | Notes |
|--------------|------------|-------|
| Claude Code | Yes | Auto-reads CLAUDE.md |
| Gemini CLI | Yes | Auto-reads GEMINI.md |
| GitHub Copilot | Yes | Reads .github/copilot-instructions.md |
| Cursor | Yes | Reads .cursorrules |
| Aider | Yes | Reads CONVENTIONS.md via config |
| ChatGPT | Yes | Upload or paste file contents |
| Any file-aware AI | Yes | Same methodology applies |

---

## Quick Start (5 Minutes)

### 1. Choose your deployment level:

| Level | Files | Best For |
|-------|-------|----------|
| **Lite** | 3 | Quick experiments, solo projects |
| **Standard** | 9 | Team projects, ongoing development |
| **Max** | 17+ | Enterprise, complex multi-phase projects |

**See:** [templates/DEPLOYMENT.md](templates/DEPLOYMENT.md) for full details

### 2. Copy templates to your project:

**Lite (minimum viable):**
```
your-project/
├── CLAUDE.md              # From templates/core/CLAUDE.md.template
├── MyApp_Session.md       # From templates/core/PROJECT_Session.md.template
└── MyApp_Tasks.md         # From templates/core/PROJECT_Tasks.md.template
```

**Adding to existing project?** See [CLAUDE.md.existing-project.template](templates/core/CLAUDE.md.existing-project.template)

### 3. Start every AI session with:

| Level | Startup Prompt |
|-------|----------------|
| **Lite** | `Read CLAUDE.md and MyApp_Session.md. Summarize current state and await instructions.` |
| **Standard** | `Read CLAUDE.md, MyApp_Session.md, and MyApp_Tasks.md. Summarize state, active tasks, and suggest next action.` |
| **Max** | `Read MyApp_Startup.md and follow its instructions.` |

### 4. End every session with:

`

[truncated…]