context continuity engine

nodestone

Your project remembers itself.

Stop re-teaching your AI every session. nodestone keeps your decisions, conventions and history in a local memory store — so your agent walks in already knowing the codebase.

storesdecisions · context · history

recallsemantic + exact

binary~3 MB (nuitka)

licenseHMAC-signed · local

runtimepython ≥ 3.11

platformwin · linux · macos

get nodestone docs

nodestone — project memorylive

$ nodestone init

✓ memory store created (~/.nodestone/project.db)

$ claude "why do we use Redis for sessions?"

▸ nodestone recall: 2 memories matched

▸ "Decided 2026-03: Redis over JWT — needed

server-side revocation. See auth ADR."

▸ answered from memory · 0 re-explaining

█

re-explaining

~3 MB

binary size

local

memory store

MCP

native server

// the problem

every session starts
from zero.

You explain the architecture. You explain why you picked Postgres. You paste the same conventions. Next session — gone. Your AI is brilliant and amnesiac. nodestone fixes the amnesia, not the brilliance.

✕re-pasting context every chat

✕the agent re-suggesting things you already rejected

✕decisions lost the moment the window closes

✕onboarding a new model = starting over

with nodestone

→ decisions written once, recalled forever

→ the agent remembers what it already tried

→ swap Claude for GPT — the memory stays

→ context survives across machines & teammates

// how it works

persistent memory

Your project's decisions, conventions and history live in a local store. Nothing to re-explain next session.

semantic recall

Ask your project anything — "why did we pick this auth flow?" — and get the answer you already wrote, months ago.

session continuity

Every chat starts where the last one ended. Your agent walks in already knowing the codebase.

local-first storage

Memory lives on your machine. HMAC-signed activation. No telemetry, no cloud sync you didn't ask for.

MCP-native

Exposes an MCP server. Claude, Cursor, any MCP host reads and writes your project memory directly.

setup

01pip install nodestone// compiled binary from PyPI

02nodestone init// browser opens · activate · memory store created

03nodestone mcp// your agent now remembers everything

// the full loop

atlas + deepstrain write solid code.
nodestone remembers the context.

Understanding, execution, and memory — one cognitive loop. atlas maps what's there, deepstrain changes it, and nodestone makes sure none of that hard-won context evaporates when the session ends.

understands

atlas

Deterministic map of your code — graph, risk, security. Knows what's there.

executes

deepstrain

Autonomous agent — reads, edits, tests, ships. Does the work.

remembers

nodestone

Persistent project memory — decisions, conventions, history. Keeps the context.

// pricing

pro

$7/mo