About SAME
Philosophy, principles, and the problem we're solving.
The mission
"Give every AI a memory. Make every developer's AI remember."
Today, AI coding assistants forget everything between sessions. You re-explain your project, your architecture, your past decisions — every single time. Context is the most valuable thing a developer produces, and right now it evaporates the moment a session ends.
SAME fixes this. It gives your AI persistent, structured memory that carries forward across sessions, across tools, across projects. Your AI picks up where it left off.
What makes SAME different
Most "memory" tools stop at storage — they save text and hope for the best. SAME is built around three pillars that turn raw storage into something your AI can actually trust and use.
SAME does three things, not one. It captures knowledge as you work — decisions, architecture, patterns. It structures that knowledge over time, consolidating duplicates and resolving contradictions. And it activates the right subset for the current task through briefing. Storage is a commodity. Trustworthy continuity is the product.
Memory that can't be trusted is worse than no memory at all. Every operation in SAME is transparent, reversible, and inspectable. Consolidation produces diffs, not opaque rewrites. Provenance tracks where knowledge came from. Stale context flags itself. You always know what your AI knows and why it knows it. Read more about memory integrity →
The AI manages its own memory. It captures knowledge, consolidates duplicates, checks vault health, and suppresses stale notes — all through MCP tools. The human never has to think about memory. You just notice your AI keeps getting better at understanding your work.
How to think about it
Brief is a briefing officer, not a librarian. It doesn't wait for you to search — it tells your AI what's relevant before you ask.
When your AI starts a new session, SAME's brief command assembles exactly the context it needs: recent decisions, relevant architecture, open questions. Think of it as a handoff from your past self to your present AI.
Consolidation is sleep for your AI's memory. It prunes noise, strengthens the important things, and resolves contradictions.
Over time, raw notes accumulate. Consolidation merges duplicates, reconciles conflicting information, and produces clean diffs you can review. The vault doesn't just get bigger — it gets better.
A vault is a garden, not a library. Libraries get bigger. Gardens get better.
Each project gets its own vault. The AI tends it autonomously — weeding stale notes, strengthening connections, checking health. You get a living knowledge base that improves with use, not a dusty archive that rots with time.
MCP is the USB port for AI tools. SAME plugs in once, works everywhere.
SAME integrates via the Model Context Protocol. That means it works with Claude Code, Cursor, Windsurf, Codex CLI, Gemini CLI — any tool that supports MCP. One memory, every tool.
Design philosophy
- Memory is a feature, not a hack. It's designed from the ground up as a first-class system, not bolted on after the fact.
- Open and portable. Works with every MCP-compatible tool. Your memory isn't locked to any single vendor or platform.
- Security is non-negotiable. Local-first by default. No telemetry. No outbound network calls. PII scanning built in. Your data stays on your machine.
- One command is the right number of commands. Install with one command. Configure with one command. Brief with one command. Complexity belongs inside the engine, not in front of the user.
Who SAME is for
- Developers who use AI coding tools daily and feel the "starts from zero" pain every session.
- People who work across multiple AI tools — Claude Code and Cursor and Windsurf — and want one memory that follows them.
- Anyone who values their data staying on their machine, not in someone else's cloud.
SAME is currently focused on individual developers and small teams. Enterprise features and non-technical user interfaces are on the roadmap but not the current priority.
The stack
SAME ships as a single binary with no runtime dependencies. Here's what's inside:
SQLite and sqlite-vec handle storage and vector search. Ollama (default), OpenAI, or any OpenAI-compatible provider for embeddings. No API keys required with Ollama. Works in keyword-only mode with no provider at all. The Model Context Protocol handles integration with AI tools. The BSL 1.1 license is source-available today and converts to Apache 2.0 automatically.
Get started
Install SAME with one command:
curl -fsSL https://statelessagent.com/install.sh | bash
Read the documentation, browse the source code, or join the Discord.