Persistent memory for AI coding

Dynamic memory system

Retenir learns from every session so you never have to repeat yourself. Your preferences, decisions, and project context carry over automatically — private and on your machine.

Get Started → View on GitHub
Memories captured
12,408
Local SQLite · FTS5 + sqlite-vec
Brains
6
Isolated per project
MCP tools
15
remember · recall · search · …
Last sync
2m
Supabase · v34 bidirectional
How It Works

Three steps, fully automatic

01

You work normally

Use your AI coding assistant like you always do. Retenir listens in the background and captures important knowledge from your conversations.

02

Knowledge is stored

Memories are organized, deduplicated, and strengthened over time. Everything stays on your machine in a local database.

03

Context comes back

Next session, the right memories are automatically surfaced. Your AI assistant already knows your preferences, past decisions, and project setup.

Without Memory

Every session starts from scratch

01

Repeated explanations

Your preferences and project setup — explained yesterday, forgotten today. Without persistent memory, you re-teach the same context every session.

02

Lost context

Decisions you made, bugs you fixed, patterns you established — all gone. There's no continuity between sessions.

03

Knowledge that should compound

Every session makes your AI assistant smarter — if it could remember. Persistent memory turns scattered conversations into compounding knowledge.

In Practice

The difference is immediate

Without Retenir
How should I implement search?
I'd be happy to help! Could you remind me what kind of app this is and what stack you're using?
It's the recipe app. TypeScript, Flask. We agreed on feature folders last week…
With Retenir
How should I implement search?
3 memories injected automatically
Based on your recipe app with Flask + TypeScript and the feature-folder structure you set up, I'd add a search/ module. Want me to start with the endpoint or the frontend component?
Features

What makes it work

01

Learns automatically

Works in the background while you code. Knowledge is captured from every session without any manual tagging, note-taking, or changes to your workflow.

02

Finds what matters

Uses multiple search strategies to surface the most relevant memories for each session. The right context shows up at the right time.

03

Gets smarter overnight

Organizes, deduplicates, and strengthens important memories automatically. Detects contradictions when your decisions change. Memory improves while you rest.

04

Private by default

Everything stays on your machine. No cloud accounts, no data uploads, no API keys required for the core experience. Your project knowledge never leaves your laptop.

05

Connects the dots

Links related concepts together into a knowledge graph. When you ask about one topic, related decisions and context surface naturally.

06

Works with your tools

Integrates directly with Claude Code through 15 specialized tools. Your AI assistant can actively remember, search, and manage its own memory.

Get Started

Ready in a few minutes

A handful of commands and you're set. No accounts to create, no services to configure.

terminal
# Download and install
git clone https://github.com/ClaudeStage/retenir.git
cd retenir && python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[all]"

# Connect to your AI assistant
retenir install   # set up automatic capture
retenir register # enable memory tools

# That's it. Retenir handles the rest.

Requires Python 3.11+ and macOS. Works best on Apple Silicon. See the full setup guide for details.

FAQ

Common questions

Is it free?
Yes, completely. Open source, no paid tiers, no usage limits. I built it for my own workflow and released it for everyone.
Does it need an internet connection?
No. Everything runs locally on your machine by default. It uses on-device AI models for processing, so the core experience works fully offline.
Is my data private?
Completely. All memories are stored locally as simple files on your computer. Nothing is sent to any server. No analytics, no telemetry. You can read, edit, or delete any memory at any time.
What AI models does it work with?
It auto-detects the best available option: your existing Claude Code session (easiest), local Apple Silicon models (private), Ollama (flexible), or the Claude API (optional). No configuration needed for most setups.
How is this different from writing notes manually?
Manual notes require effort and become outdated. Retenir learns automatically from every session, searches intelligently across all your memories, manages what to surface and when, and keeps itself up to date overnight. Think of it as notes that write and organize themselves.
Recall

Every session smarter than the last

Three commands and Claude Code starts compounding what it learns from you.

Get Started → View on GitHub