NEWOne memory for every agent, bot and chat windowOne memory for every agent and botWhat’s new →

Editor Integration

Editor plugins,
and the MCP server underneath.

SuperLocalMemory ships two ways. As a native plugin for Claude Code, Codex, Antigravity, VS Code / GitHub Copilot and Hermes, with skills, sub-agents and session hooks packaged together, plus a lighter Cursor-format plugin for Grok Bot and Cursor. And as a plain MCP server that any Model Context Protocol client can connect to directly, with no plugin required.

Native plugins

One package.
Skills, agents, commands, hooks.

The five editor plugins (Claude Code, Codex, Antigravity, VS Code / Copilot and Hermes) install the full surface at once: sixteen skills, four advisory sub-agents and session hooks, all built from one source in the repository so they stay in step. Grok Bot and Cursor get a smaller bot-sized plugin, described below the cards.

Claude Code

plugin

Installs from the qualixar marketplace. Restart Claude Code to load it after install or update.

16 skills
4 sub-agents
slash command + hooks

terminal

# adds the marketplace, then installs
claude plugin marketplace add qualixar/superlocalmemory
claude plugin install superlocalmemory@qualixar

Codex

plugin

A dedicated Codex build — its own rules, hooks, and launcher, not a relabeled copy of another editor's plugin.

16 skills
4 sub-agents
Codex-shaped hooks

terminal

# adds the marketplace, then installs
codex plugin marketplace add qualixar/superlocalmemory --ref main
codex plugin add superlocalmemory-codex@qualixar

Antigravity

plugin

Antigravity's plugin installer takes a local directory, not a remote marketplace — clone the repository first.

16 skills
4 sub-agents
session hooks

terminal

# clone, then point agy at the plugin folder
git clone https://github.com/qualixar/superlocalmemory.git
agy plugin install superlocalmemory/antigravity-plugin

VS Code / GitHub Copilot

plugin

One command from your project root writes the MCP config and the Copilot instructions file, merging without overwriting what is already there. The prompt files, advisor agents and hooks come next: copy them from the plugin's .github/ folder into your project's .github/, keeping any files of your own.

16 skills as prompt files
4 advisor agents
session hooks file (Preview in VS Code)

terminal, from the project root

# writes .vscode/mcp.json and .github/copilot-instructions.md
slm connect vscode-copilot --here
# then copy copilot-plugin/.github/ (prompts, agents, hooks)
# into your project's .github/. The npm package carries it at
# $(npm root -g)/superlocalmemory/copilot-plugin/

Hermes

plugin

A native Hermes plugin with /slm <command> slash commands, plus /slm-<command> aliases for the public SLM CLI. It installs disabled: you review it and enable it. The plugin only calls the superlocalmemory MCP server you grant it. It cannot grant itself capabilities.

16 skills
4 on-demand child-agent roles
/slm slash commands

terminal

# runtime first, then the plugin
python -m pip install --upgrade superlocalmemory
slm doctor
hermes plugins install 'https://github.com/qualixar/superlocalmemory.git#hermes-plugin'

Another editor?

slm connect <ide> writes the MCP config for your IDE, including Cursor, Windsurf, Zed, JetBrains, Gemini CLI and Claude Desktop. No plugin needed to get the memory tools.

Grok Bot / Cursor-format plugin

A different shape from the editor plugins: a Cursor-format plugin from the qualixar marketplace, built for a shared, memory-tight bot computer. It has no hooks and no dashboard. No API key and no sign-in: the server runs locally on the bot computer, so nothing goes to a memory SaaS. It is not the same thing as slm connect cursor, which only writes MCP config.

2 skills for bots sharing a computer (slm-bot-memory, slm-getting-started-bot)
core MCP profile, 19 tools
lite resource profile: about 190 MB resident, against about 390 MB with the reranker on (measured on macOS)

host Plugins screen

# in Grok Bot's (or Cursor's) Plugins screen
1. add the marketplace: qualixar
2. add the plugin: superlocalmemory

Compare

What each host
ships.

The same memory behind every row. What changes is how much of the surrounding kit your host can take.

Host	Skills	Sub-agents	Hooks	How you install
Claude Code	16	4	Session hooks, plus the `/slm-loop` command	Add the qualixar marketplace, then `claude plugin install superlocalmemory@qualixar`
Codex	16	4	Codex lifecycle hooks (review them with `/hooks`)	`slm connect codex` then `slm codex install`, or add the marketplace and `codex plugin add superlocalmemory-codex@qualixar`
Antigravity	16	4	Session hooks	Clone the repository, then `agy plugin install` on the plugin folder
VS Code / Copilot	16, as prompt files	Advisor agents (4)	A hooks file: stable on Copilot CLI and the cloud agent, Preview in VS Code	`slm connect vscode-copilot --here`, then copy the add-ons into `.github/`
Hermes	16	4 on-demand child-agent roles	Lifecycle observation; your Hermes hooks and settings are left alone	`hermes plugins install` from the repository, after the runtime is installed
Grok Bot / Cursor-format	2	Not listed	None, and no dashboard	Add the qualixar marketplace in the host's Plugins screen, then `superlocalmemory`
Any MCP client	Tools only: profiles from 9 to 110 tools	None	None	`slm connect <ide>`, or one `mcpServers` entry

The five editor plugins are built from one source and adapted to each host. The Antigravity install takes a local folder, which is why it starts with a clone.

What the plugins teach your agent

Sixteen skills.
Four agents. Hooks that fail open.

Every plugin is built from one source, so every host gets the same sixteen skills and four agents, adapted to that host.

slm-recallRecall and search: filters, tags, kinds, time windows, the answer check
slm-rememberWhat to save, memory kinds, replaces, corrections
slm-sessionSession start, pinned items and standing rules, closing a session
slm-statusHealth, slm db integrity and slm db repair, embedder and model checks
slm-governanceScopes, roles, retention, audit trail, GDPR export and erasure
slm-scopePersonal, shared and global memory across profiles
slm-profileMemory profiles and switching between them
slm-graphThe knowledge graph and entity tools
slm-meshSLM-Mesh: messages, waits, locks and shared state between agent sessions and web apps
slm-loopBounded loops and their history
slm-cacheContext optimization: the exact KV cache
slm-compressContext optimization: reversible compression
slm-web-accessSetting up and diagnosing Web access for web apps
slm-mediaPictures, PDFs and folders
slm-bot-memorySeveral bots sharing one computer and one store
slm-getting-started-botFirst steps for a bot computer

Four agents

A memory advisor, an optimize advisor, a governance advisor and a loop runner.

Commands

/slm-loop in the Claude Code, Codex and Antigravity plugins. Hermes has /slm <command> with a /slm-<command> alias for each, and Copilot carries the sixteen skills as prompt files.

Hooks

Run at session start, after tool use and at the end of a session, and every hook fails open: a hook problem never blocks your agent. Per-prompt recall and the session summary can be switched off. Recalled memory is injected only when it is a supported answer, under a token cap.

Plugins, skills and hooks guide → · Auto-memory →

MCP memory server

Pick how many tools
your agent sees.

Set SLM_MCP_PROFILE. Over stdio the command is slm mcp; over HTTP it is http://127.0.0.1:8765/mcp/.

Profile	Tools	For
`core`	19	Remember, recall, expand, sessions, optimize, correction review
`code`	39	Core plus code graph, memory kinds, Brain evidence, profile switching, bounded loops
`mesh`	9	SLM-Mesh coordination only
`full (and unset)`	58	Memory, kinds, Brain, optimize, skill evolution, mesh, loops, views and summaries (63 with pictures and documents on)
`power`	70	Full plus administration, lifecycle and diagnostics (75 with pictures and documents on)
`whole`	110	Every registered tool

MCP tools reference → · Framework adapters for nine frameworks are listed further down this page. Dedicated setup pages: Claude Code, Cursor, ChatGPT, Perplexity.

For agents

Let your agent
install it for you.

Point an agent at SETUP_FOR_AGENTS.md: numbered, non-interactive steps with the expected output of each. Install, slm init --auto, slm serve start, slm warmup, connect the host, then call session_init at the start of every session and remember and recall as you work. The site also publishes llms.txt and llms-full.txt for crawlers and agents, and the universal agent rules cover any harness not listed here.

No plugin required

Any MCP client
connects directly.

A plugin is a convenience — skills, sub-agents, and hooks bundled for editors that support the format. The memory itself is served over the Model Context Protocol, which any compliant client can reach with a single server entry.

Add the server

Works the same way in Cursor, a raw mcpServers block, or any tool that speaks MCP over stdio or HTTP — no marketplace, no plugin manager. Run slm connect <ide> and SLM writes this config for the IDE you name, including Cursor, Windsurf, Zed, JetBrains, Gemini CLI and Claude Desktop.

mcp.json

{
  "mcpServers": {
    "superlocalmemory": {
      "command": "slm",
      "args": ["mcp"]
    }
  }
}

Or point the client at the local HTTP endpoint instead of starting a process:

mcp.json (HTTP)

{
  "mcpServers": {
    "superlocalmemory": {
      "type": "http",
      "url": "http://127.0.0.1:8765/mcp/"
    }
  }
}

Full MCP client compatibility →

Framework adapters

Native memory
for nine frameworks.

Writing code against an agent framework? Nine small packages plug SLM in as the framework’s native memory store, so framework memory lands in SLM — recallable from the CLI, any MCP client or the dashboard. If you only want an AI tool to use SLM, you don’t need any of this.

Framework	Package	Install
LangChain	`langchain-superlocalmemory`	On PyPI
LlamaIndex	`llama-index-storage-chat-store-superlocalmemory`	On PyPI
CrewAI	`crewai-superlocalmemory`	On PyPI
LangGraph	`langgraph-superlocalmemory`	From a repository clone
Semantic Kernel	`semantic-kernel-superlocalmemory`	From a repository clone
Microsoft Agent Framework	`agent-framework-superlocalmemory`	From a repository clone
AutoGen	`autogen-superlocalmemory`	From a repository clone
Google ADK	`google-adk-superlocalmemory`	From a repository clone
OpenAI Agents	`openai-agents-superlocalmemory`	From a repository clone

Only three of the nine are on PyPI today; the other six install from a clone of the repository. Python 3.12 to 3.14 (CrewAI: 3.12 and 3.13). Each adapter tags its records so one conversation never mixes with another.

Any agent

One rulebook
for any agent.

The universal agent rules are a single file that teaches an agent when to recall, what to save and how to keep memory clean. The same discipline the native plugins carry, for agents that have no plugin.

Paste it where your agent reads its instructions

Drop the file into AGENTS.md, CLAUDE.md, .cursorrules or the agent's system prompt. It covers when to load context at session start, recall before remembering, one atomic fact per save with tags and a kind, and what to do when recall says it has no answer.

Read the agent rules → Architecture: the full picture →

Web access

Web apps
and bots.

ChatGPT, Claude on the web, Claude Code on the web, Muse and Composio can use the memory on your computer. Turn on Web access from the dashboard's Connected apps page; the memory database stays on your computer.

Give the app its instructions

Connecting an app gives it the tools, not the judgment to use them well. The web-agent pack closes that gap with three copy-paste shapes: an instructions block for the app's system-prompt field, an Agent Skill folder for apps that accept skills, and a Copy instructions button in the dashboard. They tell the agent to recall before answering, to say "I don't have that" when memory has no answer, to save only lasting facts and never secrets, and what to do when your computer is asleep. On your computer, the slm-web-access skill, shipped in every editor plugin, helps a local agent set up Web access and diagnose a connection.

Web-agent instructions → Architecture: the full picture →

Not just editors

Works with web apps
and bots too.

The same governed memory reaches bots and web apps. Web apps sign in with OAuth and you choose per app whether it may only read or also save. A bot on a shared computer runs the plugin locally.

Grok Bot

Runs the Cursor-format plugin on the bot computer itself, so nothing goes to a memory SaaS. Bots on one computer share one memory store by default.

No API key, no sign-in

ChatGPT

Uses the memory on your computer through Web access; paste the web-agent instructions into a project or custom GPT so it knows when to recall.

Web access · OAuth sign-in

Muse

Muse bots reach the same memory through Web access and take the same pasted instructions, so they recall before answering.

Web access · read or save, per app

Claude on the webAdd it under Connected apps and it recalls from the memory on your computer. Claude Code on the web works the same way.

ComposioComposio agents connect over Web access and take the same pasted instructions block.

ChatGPTProjects and custom GPTs have an instructions field for the web-agent block.

MuseMuse bots take the same pasted instructions, and any app can be removed from Connected apps with immediate effect.

The memory database stays on your computer, and your computer must be online for a web app to reach it. See the day-to-day flow, step by step →

Package

Read the guidesPlugins, skills and hooks · IDE setup · Web access guide · Framework adapters · SETUP_FOR_AGENTS.md (for agents)

★ Star on GitHub

Editor plugins,
and the MCP server underneath.

One package.
Skills, agents, commands, hooks.

Claude Code

Codex

Antigravity

VS Code / GitHub Copilot

Hermes

Another editor?

Grok Bot / Cursor-format plugin

What each host
ships.

Sixteen skills.
Four agents. Hooks that fail open.

Four agents

Commands

Hooks

Pick how many tools
your agent sees.

Let your agent
install it for you.

Any MCP client
connects directly.

Add the server

Native memory
for nine frameworks.

One rulebook
for any agent.

Paste it where your agent reads its instructions

Web apps
and bots.

Give the app its instructions

Works with web apps
and bots too.

Grok Bot

ChatGPT

Muse

Install the CLI

MCP integrations

Operating modes

Editor plugins,and the MCP server underneath.

One package.Skills, agents, commands, hooks.

Claude Code

Codex

Antigravity

VS Code / GitHub Copilot

Hermes

Another editor?

Grok Bot / Cursor-format plugin

What each hostships.

Sixteen skills.Four agents. Hooks that fail open.

Four agents

Commands

Hooks

Pick how many toolsyour agent sees.

Let your agentinstall it for you.

Any MCP clientconnects directly.

Add the server

Native memoryfor nine frameworks.

One rulebookfor any agent.

Paste it where your agent reads its instructions

Web appsand bots.

Give the app its instructions

Works with web appsand bots too.

Grok Bot

ChatGPT

Muse

Install the CLI

MCP integrations

Operating modes

Editor plugins,
and the MCP server underneath.

One package.
Skills, agents, commands, hooks.

What each host
ships.

Sixteen skills.
Four agents. Hooks that fail open.

Pick how many tools
your agent sees.

Let your agent
install it for you.

Any MCP client
connects directly.

Native memory
for nine frameworks.

One rulebook
for any agent.

Web apps
and bots.

Works with web apps
and bots too.