Your AI Picks Up Where It Left Off — Introducing SAME v0.12.0

Persistent memory for AI coding agents. One install, one memory, every tool.

Every AI coding session starts from zero. You re-explain your project architecture, your naming conventions, the decisions you made last week. You paste the same context into the chat. You answer the same questions. Whether you use Claude Code, Cursor, Windsurf, or any other tool — the AI has no memory of what happened before.

SAME fixes this. It gives your AI persistent memory across tools and sessions. Install it once, and your AI retains knowledge about your project, your decisions, your patterns — everywhere. Close your terminal on Friday. Open a different tool on Monday. The context is already there.

What is SAME?

SAME (Stateless Agent Memory Engine) is a single Go binary that runs on your machine. It uses SQLite for storage and vector embeddings for semantic search. It exposes 17 MCP tools that any compatible AI client can use to store, search, consolidate, and manage memory autonomously.

SAME works with Claude Code, Cursor, Windsurf, Codex CLI, Gemini CLI, and any other MCP client. It functions as an MCP memory server — a universal memory layer that sits behind all your AI tools.

As you work, your AI stores knowledge: architectural decisions, code patterns, debugging findings, project conventions. Next session — or in a different tool entirely — it retrieves what it needs. Decisions, patterns, architecture, context. All searchable, all portable, all local.

This is cross-tool AI memory. Not memory locked inside one editor. Not a cloud service that ingests your codebase. A local engine that your AI uses to remember.

The Handoff Moment

Here is the experience that makes SAME click:

You are deep in a Stripe webhook handler. You have worked through the edge cases, identified that the idempotency check still needs attention, and mapped out the next steps. Then you close your terminal. The weekend happens.

Monday morning, you open your editor. Your AI says:

"I see from the session handoff that you were working on the Stripe webhook handler. The idempotency check is still pending — the deduplication logic needs to handle the race condition on concurrent retries. Want me to continue where you left off?"

No re-explaining. No warmup. No pasting previous conversations. Your AI is already oriented — because SAME gave it the memory it needed.

This works across tools too. Start a task in Claude Code, pick it up in Cursor. The memory follows because it lives in SAME, not inside any single tool.

What's New in v0.12.0: Memory Integrity

Storing memory is table stakes. The hard question is: can you trust it?

Code changes. Files get refactored. Decisions get reversed. If your AI agent memory is just a pile of notes with no connection to reality, you end up with confident AI that is confidently wrong — acting on stale knowledge about code that no longer exists.

SAME v0.12.0 introduces memory integrity — a system that tracks where knowledge came from and tells you when it might be outdated.

Run same health to see the state of your vault at a glance:

terminal
$ same health

Vault          /home/dev/projects/myapp/.same/vault.db
Schema         v8
Notes          247
Graph nodes    183
Graph edges    412

Trust breakdown
  validated      189  (76.5%)
  unknown         41  (16.6%)
  stale           14  ( 5.7%)
  contradicted     3  ( 1.2%)

Source files tracked   94
Diverged sources       7
Embeddings coverage   100%

Vault is healthy. 7 sources have diverged — run same integrity check for details.

"Everyone builds memory storage. We built memory you can trust."

How It Works

Getting started takes about 30 seconds.

1. Install

terminal
$ curl -fsSL statelessagent.com/install.sh | bash

This downloads the SAME binary for your platform. Single binary, no dependencies, no runtime.

2. Initialize

terminal
$ same init
Detected project: myapp (Go)
Created vault: .same/vault.db
Installed Claude Code hooks
Configured MCP server
Ready. Your AI now has persistent memory.

same init detects your project, creates a local vault, and configures MCP integration for your tools. Your AI now has access to 17 MCP tools for searching, storing, consolidating, and managing memory.

3. Try it

terminal
$ same demo
Loading sample vault with 50 notes...
Starting MCP server on stdio...
Demo ready. Open Claude Code or Cursor and ask about the project.

same demo loads a sample vault so you can see how cross-tool AI memory feels before using it on your own codebase.

Who Is This For?

Try SAME

Persistent memory for AI coding agents. Local, private, works everywhere.

terminal
$ curl -fsSL statelessagent.com/install.sh | bash
$ same demo