v1.0.15 — Project switching fixed

What's fixed

Bug: Wrong project data after handshake / search stuck on wrong repo

Before: After handshake(project_path="C:\Dev\my-project"), subsequent search_repo() and build_context() calls silently switched back to the MCP server's working directory (e.g. System32 or AtlasMemory itself). Reported and confirmed by Codex.

After: Once a project is explicitly bound via project_path, the session stays locked to that project for all tool calls.

Changes

• handshake(project_path=...) now locks the session — cwd auto-detect no longer overrides it
• index_repo(path=...) also sets the session lock
• Handshake brief uses the bound project root (not process.cwd())
• ensureIndexed(), generate_claude_md, migrateSessionHandoff all use bound root

Validated

• 3-repo switch test: claude-learn → AtlasMemory → call-recorder, all tool calls stay on correct project
• Confirmed by Codex: handshake(project_path=...) now returns correct project (1892 files) with no System32 bleed-through
• 111 unit tests passing

Usage

To bind to your project at session start:

handshake(project_path="/path/to/your/project")

All subsequent tool calls (search_repo, build_context, analyze_impact, etc.) will use that project.

v1.0.14 — Claude CLI compatibility + Windows path fix

Fixes

Critical: Global DB pollution eliminated

• resolveProjectRoot() no longer falls back to ~/.atlasmemory — uses ephemeral temp dir instead
• Prevents cross-project data mixing that caused wrong handshake results
• isValidProjectDir() now rejects .atlasmemory paths

Claude CLI enrichment fix

• Removed deprecated --max-tokens flag (no longer supported by Claude CLI)
• atlas enrich now works cleanly with latest Claude Code

VS Code extension v0.1.6

• Fixed Windows "Program Files" path with spaces crash ('C:\Program' is not recognized)
• Proper path quoting when shell: true

MCP improvements

• handshake tool now accepts project_path parameter for explicit workspace binding
• ensureIndexed() accepts optional path for project routing
• Warns on stderr when stuck on orphan/fallback DB

Upgrade

npm install -g atlasmemory@1.0.14

v1.0.12 — Version reporting fix + all previous fixes

Fixes

• Version reporting — --version was stuck at 1.0.0 in the bundle. Now correctly shows 1.0.12.
• Includes all fixes from v1.0.9-v1.0.11 (project isolation, generate writes to disk, auto-detect workspace)

How to update

npm install -g atlasmemory@latest

Then restart your AI tool (Codex, Claude, Antigravity) to pick up the new MCP server.

v1.0.11 — Auto-detect workspace on every tool call

Critical Fix

Auto-detect workspace on every MCP tool call — fixes the #1 user-facing bug where AtlasMemory would bind to the wrong project.

What was broken

When an AI tool (Codex, Antigravity, etc.) called handshake() or search_repo() without first calling index_repo, the MCP server stayed on whatever DB it started with — often a stale global DB or a different project entirely.

What's fixed

ensureIndexed() now checks if the current working directory has a local .atlas/atlas.db and automatically switches to it. This runs on every tool call, so switching between projects just works.

Reported by

OpenAI Codex (GPT-5.4) stress test — 10 bugs traced to this single root cause.

Full Changelog

v1.0.10 → v1.0.11: v1.0.10...v1.0.11

v1.0.10 — Project isolation + generate writes to disk

What's New

• v1.0.10 — generate_claude_md now writes files to disk by default via MCP (previously only returned content)
• v1.0.9 — All intelligence services (enrichment, impact, memory, learner) now rebuild on project switch — fixes enrichment targeting wrong project

Bug Fixes

• Fixed: enrich_files was enriching wrong project's files when switching between projects
• Fixed: generate_claude_md via MCP only returned content without writing CLAUDE.md/.cursorrules to disk
• Moved global .atlas/ DB to project-local — each project now has isolated memory

Full Changelog

v1.0.8 → v1.0.10: v1.0.8...v1.0.10

v1.0.8 — Project-local DB fix (no more global store)

What's New

• Project-local .atlas DB — MCP server now creates .atlas/atlas.db in each project directory instead of a single global DB
• Fixes VS Code extension showing "Not Indexed" when MCP already indexed
• Fixes incremental index not working (was always full re-index via MCP)
• Fixes multiple projects sharing one DB (data mixing)

How it works now

project-a/.atlas/atlas.db  ← project A's index
project-b/.atlas/atlas.db  ← project B's index (isolated)

When index_repo(path) is called, AtlasMemory switches to that project's local DB automatically.

Full Changelog

v1.0.7 → v1.0.8: v1.0.7...v1.0.8

v1.0.7 — Full i18n sync + enrichment docs + enterprise monorepo

What's New in v1.0.7

• All 6 i18n READMEs fully synced with English (Maximum Power Checklist, enrichment tables, one-time cost note)
• Enrichment token cost warning in CLI dry-run
• Default generate = all (5 AI instruction files at once)
• Windsurf + Antigravity generate support
• Enterprise monorepo tested on Next.js (28K files, 3522 indexed, zero crash)
• Coolify fix — no more segfault on large PHP/minified JS repos

Tested At Scale

Project	Files	Symbols	Time	Status
Small (1 file)	1	1	200ms	Zero crash
Express (141 files)	141	124	1s	Zero crash
Coolify (1442 PHP)	1442	6831	12s	Zero crash
Next.js (28K total)	3522	34,579	6 min	Zero crash

Generate All Formats (NEW default)

npx atlasmemory generate   # Now produces ALL 5 formats:
# CLAUDE.md, .cursorrules, copilot-instructions.md, .windsurfrules, AGENTS.md

Full Changelog

v1.0.0 → v1.0.7: v1.0.0...v1.0.7

v1.0.5 — Enterprise monorepo support (Next.js 28K files)

What's New

• Enterprise monorepo support — tested on Next.js (28K files, 3522 indexed, zero crash)
• Smart directory exclusion — test, fixture, docs, bench, examples auto-excluded
• 5000 file safety limit with warning message
• Enrichment token warning — shows estimated token cost before enriching large projects

Tested At Scale

Project	Files	Symbols	Time	Status
Small (1 file)	1	1	200ms	Zero crash
Express (141 files)	141	124	1s	Zero crash
Coolify (1442 PHP)	1442	6831	12s	Zero crash
Next.js (28K total)	3522	34,579	6 min	Zero crash

Full Changelog

v1.0.4 → v1.0.5: v1.0.4...v1.0.5

v1.0.4 — 5-platform AI instructions + enterprise-scale fixes

What's New in v1.0.4

• Default generate now produces all 5 formats — CLAUDE.md, .cursorrules, copilot-instructions.md, .windsurfrules, AGENTS.md
• Windsurf + Antigravity support — generate --format windsurf and --format antigravity
• Enterprise-scale fix — Tested on Coolify (1442 PHP files, zero crash)
• OOM prevention — Minified file detection (>2000 char lines), vendor/public exclusion, 512KB limit

Quick Start

npx atlasmemory demo
npx atlasmemory index .
npx atlasmemory generate        # Produces 5 AI instruction files
npx atlasmemory enrich --all

MCP Setup

{ "mcpServers": { "atlasmemory": { "command": "npx", "args": ["-y", "atlasmemory"] } } }

Full Changelog

v1.0.0 → v1.0.4: v1.0.0...v1.0.4

v1.0.0 — Proof-backed AI memory for your codebase

AtlasMemory v1.0.x

Turn any codebase into an AI-readable neural map — with proof.

Latest: v1.0.4

Highlights

• 🔍 11 Language Support — TypeScript, JavaScript, Python, Go, Rust, Java, C#, C, C++, Ruby, PHP
• 🛡️ Proof System — Every claim linked to code anchors (line + SHA-256 hash)
• 📦 Token-Budgeted Context — Greedy-optimized packs for any context window
• 🤖 28 MCP Tools — Claude, Cursor, Copilot, Windsurf, Antigravity
• ✨ AI Enrichment — Semantic tags via Claude CLI or OpenAI Codex
• 🏗️ 5-format AI Instructions — generate produces CLAUDE.md, .cursorrules, copilot-instructions.md, .windsurfrules, AGENTS.md
• 🧠 Cross-session Learning — Decisions persist across sessions
• 💥 Enterprise-scale — Tested on Coolify (1442 files, zero crash)

Quick Start

npx atlasmemory demo
npx atlasmemory index .
npx atlasmemory generate
npx atlasmemory enrich --all

MCP Setup (Zero Config)

{ "mcpServers": { "atlasmemory": { "command": "npx", "args": ["-y", "atlasmemory"] } } }

Changelog since v1.0.0

• v1.0.4 — Default generate format is now all (5 files at once)
• v1.0.3 — Added Windsurf (.windsurfrules) + Antigravity (AGENTS.md) support
• v1.0.2 — Fixed segfault on large repos (Coolify 1442 files, minified JS)
• v1.0.1 — Fixed OOM on repos with vendor/public directories

Links

• 📦 npm
• 🧩 VS Code Extension
• 🌐 Glama.ai