One shared memory across every AI tool

YourAItools
forgeteverything
whenyouswitch

You explain your project to Claude, then open Cursor and explain it all over again. Samskriti keeps one shared memory — decisions, tasks, bugs — so whatever tool you're in already knows what's going on. Local, open source, runs on your machine.

Live now · Free · Open-source · MIT licensed

Get Started Free View on GitHub

Scroll

Install in two minutes

How to Install.

Install it, run one command, restart your tool. That's it — no config files to mess with, and nothing leaves your machine.

Install with pipx

pipx install git+https://github.com/Escalate17/samskriti-project

Don't have pipx? Run python3 -m pip install --user pipx && python3 -m pipx ensurepath, then restart your terminal.

Connect it — one command

samskriti-project setup

This writes the MCP config for you (and backs up any existing config). Use --cursor, --codex, or --all for other tools. Prefer to edit config by hand?

Restart your AI tool

Fully quit and reopen Claude Code, Cursor, or Codex. A running session won't pick up the server — MCP servers load only when the tool starts.

Verify the connection

In Claude Code, type /mcp — you should see samskriti-project listed with its 8 tools.

Helper commands

Three shortcuts for the most common actions. Trigger them with /sam plus plain English — there is no standalone /catchup command.

catchup“catch me up” — recap of the latest entries plus how many tasks are open.
open“what's open” — lists the active tasks.
logquick-log a decision (title auto-derived from the text).

Claude Code / Cursor

/sam catch me up
/sam what's open
/sam log dropping the Redis cache

Any tool (plain English)

use samskriti-project
  to catch me up

Full documentation on GitHub

See it work

One Decision, two Tools.

Record something in one tool, read it back in another. No copy-paste, no re-explaining. Here's the actual handoff:

Claude Code recording a project decision through Samskriti — 1Claude Code— Records a decision into the shared project memory.

Same shared memory

Gemini reading back a decision recorded by claude-code via Samskriti — 2Gemini— Reads the same decision back — attributed to claude-code.

Deterministic & Auditable

Distilled State.
Zero Conversation Logs.

Samskriti doesn't store or upload your chat logs. It pulls out the decisions and rules and keeps them as plain JSON on your machine. Nothing to audit, nothing in the cloud, easy to move.

0chat messages logged

100%local-first runtime

JSONportable state schemas

Pricing

Free and Open-Source.

The local runtime is free, MIT licensed, and yours to run. No account, no cloud.

Local · Open-source

Free

Live now — MIT licensed, runs on your machine

✓Local MCP server — runs entirely on your machine
✓One shared project memory across every AI tool
✓5 tools: record, read, list, search, update
✓Unlimited projects and entries
✓MIT licensed — open-source, no account, no cloud

Get Started Free

Where this is going

A Shared Memory Layer
is Just Step One.

A shared memory layer is step one. The goal is an AI that carries your context, your style, and your history across every tool you use — one continuous companion, not a dozen strangers.

Conversation

Session begins

Experience

Context analyzed

Samskriti

Behavior updated

Growth

State persisted

Next Chat

Resumes with memory

Architecture

An External Behavioral Layer

Samskriti runs outside the model. The LLM handles the reasoning, while the Samskriti runtime governs and persists behavioral state.

User Input

Prompt / Event

LLM

Reasoning (Stateless)

Samskriti Runtime

Behavioral State

The goal

More Than Memory. A Companion That Carries you.

Standard model memory recalls text facts. This is the direction we're building toward — most of these are goals, not features that ship today. Shared project memory across tools is the first step, live now above.

Where this is going	ChatGPT / Claude Memory	The Samskriti goal
Remembers facts you tell it		The goal
Carries project context across different AI tools		The goal
Grows a personality over time		The goal
Picks the relationship back up where you left off		The goal
Keeps its personality after a model upgrade		The goal
Sticks to the rules you set, reliably		The goal
Snapshot and roll back its state anytime		The goal
Runs entirely on your own machine		The goal

A normal AI vs. Samskriti

Same Conversation.
One Actually Remembers.

A normal AI

With Samskriti

Picking up where you left off

Hi! How can I help you today?

A normal AI: May recall a few saved facts, but resets its tone and forgets how you actually like to talk every new chat. You re-explain yourself constantly.

remembers & evolves

Hey, back to chapter 3 of your novel — want to keep the darker tone we landed on last night?

With Samskriti: It carries your shared history and the personality it has grown into, so it continues the relationship instead of starting over.

When you upgrade the model

You're on a new model now. Hi! How can I help you today?

A normal AI: Swap to a newer model and its behavior resets — the voice you got used to is gone, and you train it on your preferences from scratch.

Upgraded my engine under the hood — still me. Same voice, same memories, same in-jokes.

Realizing memory is not the same as a continuous personality

Its personality lives outside the model, so it survives the upgrade.

Staying consistent over a long chat

Sure, I'll keep it short! …and then writes six long paragraphs anyway.

A normal AI: The longer the conversation runs, the more it drifts and forgets the boundaries you set earlier.

Keeping it to three lines, like you asked — every time.

The model and Samskriti working together

Your rules are kept as real state — not a fading line in the prompt.

Tried it?

Would you Pay for a Hosted Version?

Tried it? One question: would you pay for a hosted version you don't have to install or run yourself? Drop your email if you want updates.

YourAItoolsforgeteverythingwhenyouswitch