Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Nexus — Persistent AI Memory

    3

    Equip your agent with a persistent, local Markdown-based "second brain" for long-term project memory.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    5.0 (1 review)
    Nexus — Persistent AI Memory

    Nexus — Persistent AI Memory

    Example session with this skill installed

    I decided to switch the API to use Bearer tokens instead of API keys for the dev portal. Can you remember that for next time?

    • Read your context and instructions
    • Compiled the nexus

    I have searched your vault for existing notes on the dev portal authentication and updated your records. I created a new decision note that supersedes the previous "API Key Auth" strategy to maintain the project lineage.

    Memory Update: Dev Portal Authentication

    FieldValue
    Typedecision
    Projectdev-portal
    Statusactive
    ContentSwitch authentication mechanism from static API keys to Bearer tokens (JWT) to improve security and allow for easier token rotation.

    Next steps

    1. Ask me to reindex if you edit the Markdown files directly.
    2. Ask me to "link this to the OAuth2 research note" if you want to connect these topics.
    3. Mention "the dev portal" in any future session to have me recall this decision.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Standard LLM context windows are fleeting. You lose project history, architectural decisions, and "hard-won" insights the moment a chat ends or a context limit is hit. This skill integrates a durable, local Markdown memory vault directly into your agent's reasoning loop.

    What it does

    • Proactive recall identifies when a user mention of a project or past decision warrants a search of the local vault.
    • Durable storage captures and persists decisions, principles, and project notes across completely separate chat sessions.
    • Note versioning manages the lineage of your thinking by superseding old decisions rather than simply overwriting them.
    • Contextual linking connects related notes to build a "second brain" that grows with your projects.
    • Status tracking labels notes as active, archived, or superseded to prevent the agent from citing outdated information.

    How it works

    1. Initialize the local Nexus MCP server by reindexing your existing Markdown vault folder via the CLI.
    2. Search triggers automatically when you mention specific projects or past work, pulling relevant note snippets into the current context.
    3. Read operations fetch the full content of notes to ensure the agent understands the nuances of past decisions before citing them.
    4. Upsert allows the agent to save new insights or revise existing ones based on the current conversation.

    Frameworks & tools

    Requires Node.js 22+, git, and the nexus-brain-mcp package. Works with any MCP-compatible host including Claude Code, Claude Desktop, and Cursor.

    Why this beats prompting it yourself

    Generic prompts cannot access local files across sessions without a structured retrieval system. This skill provides the specific logic needed to manage note states, handle "superseded" logic, and maintain a high bar for what is worth "remembering" to avoid context bloat.

    Use cases

    • Tracking architectural decisions for a long-term software project.
    • Recalling specific configuration choices made in previous debugging sessions.
    • Managing a personal knowledge base of engineering principles and "lessons learned".
    • Resuming complex workflows where the "state" lives in scattered Markdown notes.

    Known limitations

    Requires a local git installation to function. Does not support Intel-based Macs, Windows on ARM, or Alpine Linux. Requires approximately 1.1 GB of disk space for models and dependencies.

    Requires Node.js 22+, git on PATH, and ~1.1GB disk. No support for Intel Macs, Windows on ARM, or Alpine/musl. Requires initial manual bootstrap via CLI.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    5.0
    1 review
    5
    1
    4
    0
    3
    0
    2
    0
    1
    0

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Requires

    GitHub
    Nexus Brain MCP
    Node.js 22+
    git
    Node.js
    macOS

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files

    Allowed Hosts

    github.com

    File Scopes

    **/*.md

    Nexus reads and writes Markdown notes in your local vault, runs git to version them, and makes one outbound call to download the embedding model (once). Nothing leaves your machine otherwise. No telemetry, no cloud.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    This skill acts as a durable 'second brain' by storing and retrieving project history, architectural decisions, and lessons learned in a local Markdown vault, preventing valuable insights from being lost when chat contexts reset.

    Nexus Memory is compatible with any MCP-capable host, including Claude Desktop, Claude Code, and Cursor, provided you are running Node.js 22+ and have git installed.

    It’s mostly hands-off. Install the nexus-brain-mcp package, then just tell your agent to set it up — the skill walks it through indexing your Markdown folder and connecting the server, so you don’t have to run the steps manually. Point it at an existing vault or an empty folder to start fresh.

    Unlike basic file reading, this skill tracks each note’s state — active, archived, or superseded — so your agent never acts on a decision you’ve already revised. When you change your mind, the old note is kept and linked to its replacement, preserving the full reasoning trail instead of silently overwriting it.

    It is not compatible with Intel-based Macs, Windows on ARM, or Alpine Linux; additionally, you will need approximately 1.1 GB of free disk space for the required models and dependencies.

    New in LLM Ops