Docs

One brain. Every AI.

Memstalker gives Claude, Codex, Gemini, and ChatGPT a shared persistent memory. What one agent learns, every other agent can retrieve without you re-explaining the project.

Claude Code

Codex CLI

Gemini CLI

ChatGPT

REST API

Memstalker

Shared knowledge base

SQLite + FTS5 + semantic search + session capture

MCP connectionsClaude Desktop

memstalkerConnected

filesystemConnected

githubConnected

Custom GPT actionChatGPT

Action URL

https://api.memstalker.com/openapi.json

Auth

Header: X-API-Key

My Brain GPT is connected to your shared memory.

Overview

What is Memstalker?

It is persistent memory infrastructure for agents. Not a read-only RAG index. Not one model's built-in memory. A writeable brain your tools share.

AI sessions start stateless. Yesterday's fix disappears unless you paste it back in.

Memstalker gives every tool one shared knowledge base they can search, update, and synthesize.

That creates a self-learning loop: mistakes become reusable fixes instead of repeated context dumps.

Overview

How it works

The system keeps hot, warm, and cold memory layers so agents get fast summaries first and deeper source material only when needed.

Capture

Store sessions, fixes, and docs as soon as work happens.

Classify

Promote useful artifacts into searchable operational knowledge.

Synthesize

Generate concise briefings so agents read summaries before full docs.

Retrieve

Return the right snippet, context pack, or document when the next session starts.

Getting Started

Choose a setup path

You can start managed, self-hosted, or through ChatGPT actions. The memory model stays the same.

Fastest path

Hosted

Create an account, connect your agent with the provided config, and let Memstalker manage the infrastructure.

Free forever

Self-hosted

Clone the repo, run setup, and keep the full feature set on your own VPS or workstation.

REST API route

Custom GPT

Import the OpenAPI spec, set X-API-Key auth, and let ChatGPT search and write to the same KB.

Connect Your AI

Claude Code

The shortest path is still one command. If you want manual setup, use the MCP block below.

kb register

Registers Memstalker in Claude Code with the right stdio command.

{
  "mcpServers": {
    "memstalker": {
      "command": "npx",
      "args": ["knowledge-base-server", "mcp"]
    }
  }
}

Connect Your AI

Codex and Gemini

Both support the same stdio MCP pattern. Point them at the Memstalker server and they share the same brain immediately.

{
  "mcpServers": {
    "memstalker": {
      "command": "npx",
      "args": ["knowledge-base-server", "mcp"]
    }
  }
}

Connect Your AI

ChatGPT via Custom GPT

ChatGPT connects over REST instead of MCP, but it still reads and writes the same knowledge base.

01Create a Custom GPT in ChatGPT.
02Add an Action and import your Memstalker OpenAPI spec.
03Set API Key auth using the X-API-Key header.
04Test search and capture endpoints against your knowledge base.

Reference

MCP tools

These are the core tools exposed to MCP clients. Exact availability depends on deployment mode.

Tool	What it does
kb_search	Full-text search with ranked snippets.
kb_search_smart	Hybrid keyword and semantic retrieval.
kb_context	Token-efficient context briefings.
kb_get	Fetch a document by ID.
kb_list	List recent or matching documents.
kb_capture_session	Store a debugging or implementation session.
kb_capture_fix	Store a concise bug fix record.
kb_capture_web	Capture an external page into the KB.
kb_ingest_file	Ingest a local file.
kb_ingest_dir	Ingest a directory recursively.
kb_related	Find connected documents.
kb_update_doc	Update document metadata or body.
kb_delete_doc	Delete a document intentionally.
kb_stats	Inspect collection and search stats.
kb_health	Check server health.
kb_promote	Promote useful material into long-term memory.

Reference

REST API

ChatGPT and any non-MCP client can talk to Memstalker over HTTP using API key auth in the X-API-Key header.

Method	Endpoint	Purpose
GET	/api/v1/search?q=query	Keyword search
GET	/api/v1/search/smart?q=query	Hybrid smart search
GET	/api/v1/context?q=query	Summary-first context pack
GET	/api/v1/documents	List documents
GET	/api/v1/documents/:id	Fetch one document
POST	/api/v1/ingest	Ingest content
POST	/api/v1/capture/session	Capture a work session
POST	/api/v1/capture/fix	Capture a bug fix
POST	/api/v1/capture/web	Capture a page from the web

Reference

CLI commands

The CLI handles local setup, ingestion, search, and maintenance. Most teams only need a small subset to get started.

kb setup

kb start

kb stop

kb mcp

kb register

kb ingest

kb search

kb status

kb classify

kb summarize

kb capture-session

kb capture-fix

kb safety-check

Reference

Supported file types

Memstalker is optimized for text-heavy operational knowledge. Binary handling stays narrow on purpose.

Type	Extensions	Typical use
Markdown	.md, .mdx	Docs, notes, session captures
Plain text	.txt	Raw notes and exports
Code	.ts, .tsx, .js, .py, .go, .rs, .sh	Source files for retrieval
Config	.json, .yaml, .toml, .env.example	Tooling and deployment config
PDF	.pdf	Reference material and reports