SuperLocalMemory

Name: SuperLocalMemory
Author: Qualixar

Local, governed long-term memory shared by coding agents, with recall that says when it has no answer

By Qualixar Python Local · stdio, Streamable HTTP Knowledge and memory Developer tools

GitHub npm PyPI Website

61 popularity score ★ 274 ↓ 838/wk

Connect your client

Local (once it is running)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio Add to Goose

Opens the app, which asks before it adds the server.

Or add it by command or config

Claude CodeTerminal

claude mcp add --transport http superlocalmemory http://127.0.0.1:8765/mcp/

Cursor~/.cursor/mcp.json

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

VS Code.vscode/mcp.json

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

Codex~/.codex/config.toml

[mcp_servers.superlocalmemory]
url = "http://127.0.0.1:8765/mcp/"

Gemini CLITerminal

gemini mcp add --scope user --transport http superlocalmemory http://127.0.0.1:8765/mcp/

Local (npm)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio

Opens the app, which asks before it adds the server.

Or add it by command or config

FirstTerminal

npm install -g superlocalmemory
slm setup

Claude CodeTerminal

claude mcp add superlocalmemory -- slm mcp

Claude Desktopclaude_desktop_config.json

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

Cursor~/.cursor/mcp.json

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

VS Code.vscode/mcp.json

{
  "servers": {
    "superlocalmemory": {
      "type": "stdio",
      "command": "slm",
      "args": [
        "mcp"
      ]
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.superlocalmemory]
command = "slm"
args = ["mcp"]

Gemini CLITerminal

gemini mcp add --scope user superlocalmemory slm -- mcp

Local (PyPI)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio

Opens the app, which asks before it adds the server.

Or add it by command or config

FirstTerminal

pipx install superlocalmemory
slm setup

Claude CodeTerminal

claude mcp add superlocalmemory -- slm mcp

Claude Desktopclaude_desktop_config.json

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

Cursor~/.cursor/mcp.json

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

VS Code.vscode/mcp.json

{
  "servers": {
    "superlocalmemory": {
      "type": "stdio",
      "command": "slm",
      "args": [
        "mcp"
      ]
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.superlocalmemory]
command = "slm"
args = ["mcp"]

Gemini CLITerminal

gemini mcp add --scope user superlocalmemory slm -- mcp

Settings 4

`SLM_MCP_PROFILE`	Optional: Tool set shown to the client: core, code, mesh, full (default), power or whole.
`SLM_MCP_ALL_TOOLS`	Optional: Set to 1 to expose all 110 tools.
`SLM_MCP_TOOLS`	Optional: Comma-separated list of exactly the tools to expose.
`SLM_SKIP_LAYA`	Optional: Set to 1 to skip installing the on-device answer-check model on Apple Silicon.

What it does

SuperLocalMemory (SLM) keeps a memory store on your own computer that many AI agents can share: Claude Code, Codex, Cursor, Copilot, Claude Desktop and any other MCP client. An agent saves facts, decisions and notes with `remember` and gets them back with `recall`, `search` and `expand`. Recall answers in three tiers: a supported answer, up to three labelled closest matches, or a plain statement that nothing stored answers the question. Facts carry when they happened and when they were learned, so an agent can ask what was true at an earlier date, and corrections replace a fact while keeping it undoable.

Beyond plain memory it adds sessions, nine memory kinds (decision, rule, how-to and so on), saved views, learned behaviour patterns, skill tracking, output compression and caching, bounded agent loops, and SLM-Mesh, which lets agent sessions on one or several machines message each other, share state and take advisory file locks. A code-graph feature indexes a local repository for callers and change impact. With pictures and documents turned on, it indexes images and PDFs locally and finds them by description.

The server registers 110 tools but shows a client 58 by default (the `full` set; 63 with pictures and documents on). `SLM_MCP_PROFILE` picks a smaller or larger set (`core` 19, `code` 39, `mesh` 9, `power` 70, `whole` 110), and the docs say an unknown value is an error rather than a silent fallback (mcp-tools.md, profiles.py at v4.2.6). It runs over stdio (`slm mcp`) or as one shared HTTP daemon at 127.0.0.1:8765. In Mode A no language model is called; Mode B uses a local model through Ollama and Mode C a cloud provider you choose.

What it needs

Runtime	Node.js 18+orPython 3.12+

Node 18+ for the npm package (it creates its own Python virtual environment), or Python 3.12 to 3.14 for pip/pipx. Prebuilt installs for Apple Silicon macOS, 64-bit Windows and 64-bit Linux; none for Intel Macs or 32-bit Windows. Run `slm setup` once (pick Mode A to keep everything local). Pictures and documents need about 1.5 GB of models and 16 GB of RAM. Mode B needs Ollama or a local OpenAI-compatible server; Mode C and the online answer check (Jev) need your own provider key.

Limits

A client sees 58 of the 110 tools unless `SLM_MCP_PROFILE`, `SLM_MCP_ALL_TOOLS=1` or `SLM_MCP_TOOLS` is set; the set is fixed when the server starts. Most code-graph tools need the `whole` set, and code-graph `repo_path` must be inside your home directory. The on-device answer check (Laya) runs only on Apple Silicon; elsewhere only the online Jev check is offered. Mesh messages are capped at 4 KB and expire after 48 hours. The docs note `profile_id` is a namespace, not a security boundary. Licence is AGPL-3.0-or-later, with a separate commercial licence offered.

Who it suits

Suits developers who use several coding agents and want one memory between them that stays on their machine, with no API key needed in Mode A. Teams that need roles, audit trails and GDPR export or erasure on agent memory get those built in. It is a large system (Python runtime, background models, a dashboard, 110 tools), so someone who wants a small single-purpose memory tool with a handful of tools, or who uses an Intel Mac, may want something lighter.

Tools 58 of 110

remember: Save a memory; facts and entities are extracted and indexed
recall: Find memories for a question, with tiered support and token caps
search: Search stored memories
fetch: Fetch memories by id
expand: Drill into one recall result for more context
list_recent: List the most recent memories
update_memory: Edit a stored memory
delete_memory: Delete a memory
forget: Remove memories matching a target, with a dry-run option
get_memory_summary: Summaries of a session, day or project citing memory ids
run_view: Run a saved recall query, or list saved views
manage_view: Create, rename or delete a saved view
review_correction: Review, apply or roll back a proposed correction
list_corrections: List pending and past corrections
set_memory_kind: Set or confirm one memory's kind
memory_kinds_status: Counts per memory kind and classifier status
review_memory_kinds: Show suggested kinds awaiting review
confirm_memory_kinds: Accept suggested memory kinds
session_init: Start a session and load relevant context for a project
observe: Pass content for the server to consider saving
close_session: End a session
report_feedback: Report whether a recalled memory helped
report_outcome: Report the outcome of a task that used memories
log_tool_event: Log a tool call event for learning, with scrubbed summaries
get_assertions: List learned behavioural assertions
reinforce_assertion: Raise an assertion's confidence
contradict_assertion: Lower an assertion's confidence, deleting it below 0.2
get_soft_prompts: Return learned soft prompts for new sessions
skill_health: Invocation counts and status per agent skill
skill_lineage: Show how a skill has changed over time
evolve_skill: Run skill revision for one skill, when enabled
get_status: Counts, mode, profile and database size
set_mode: Switch between modes A, B and C
switch_profile: Change the active memory profile for the machine
run_maintenance: Run decay, pattern mining and lifecycle jobs
consolidate_cognitive: Cluster cold and archived memories into gist summaries
get_brain_evidence_status: Totals of recorded evidence receipts
record_agent_experience: Record a validated agent experience receipt
record_cognitive_turn: Record a validated receipt for one agent turn
finalize_cognitive_turn: Close a recorded turn with its outcome
observe_bounded_loop_evidence: Import read-only evidence from an installed Bounded Loops
slm_loop_run: Run a bounded loop that ends only when its gate passes
slm_loop_history: List past bounded-loop runs
slm_loop_show: Show one bounded-loop run
slm_compress: Shrink large output reversibly, keyed by id
slm_retrieve: Get the original of compressed output back
slm_cache_set: Cache a result with a time to live
slm_cache_get: Read a cached result
slm_optimize_stats: Report compression and cache savings
mesh_summary: Register this session on the mesh and say what it is doing
mesh_peers: List active peer sessions
mesh_send: Message a peer, a project or all peers (up to 4 KB)
mesh_inbox: Read unread mesh messages
mesh_wait: Wait up to 20 seconds for a new message
mesh_state: Get or set shared non-secret state
mesh_lock: Query, take or release an advisory file lock
mesh_events: Recent mesh events
mesh_status: Mesh broker health and statistics

Popularity

61 popularity score, on our list (the bar is 36)

★ 274 ↓ 838/wk

GitHub stars	274	29.3 points
Stars in 30 days	+44	8.3 points
npm downloads, last week	838	23.4 points
PyPI downloads, last month	1,807

Measured 2026-11-08. How the score works.

Where it is published

Code	qualixar/superlocalmemory ★ 274
npm	superlocalmemory 4.2.13, 2026-10-28, the one the command runs
PyPI	superlocalmemory 4.2.7, 2026-10-15

What it can reach

Runs	On your machine
Reaches	Files on your machine, A database, Your code, The internet
Changes things	Yes: it can write, send, create or delete
Sign-in	None
Licence	AGPL-3.0-or-later (npm)
Last code change	2026-10-28

Runs on your machine; memories live in a local SQLite store. Tools write, edit and delete memories, change modes and profiles, and can index local files, PDFs, images and repositories you point them at. Mode A makes no model-provider call. Data leaves the machine only through features you turn on: Mode C or cloud embedders (memory text), the Jev check (a question and the top three memories), GitHub or Google Drive backup (encrypted files), mesh peers, and Web access, which lets ChatGPT or Claude on the web call a subset of tools after a per-app sign-in. The HTTP daemon listens on 127.0.0.1:8765. Mesh messages from other agents are labelled as data; `profile_id` is not a security boundary.

Over time

GitHub stars 274 +44 (+19 %)

274230

10 Oct 20268 Nov 2026

npm downloads a week 838 −344 (−29 %)

3.5k267

16 Feb 20268 Nov 2026

Before the site measured it, read from npm's daily download counts.

Popularity score 61 +33 (+114 %)

6328

10 Oct 20268 Nov 2026

What is rising across the directory

Sources

Page checked 2026-10-15.

Badge for your README

Shows this server's rank in its category, and stars for a server that runs on your machine. It updates every hour.

MarkdownREADME.md

[![SuperLocalMemory on EveryMCPServer](https://everymcpserver.com/badge/superlocalmemory.svg)](https://everymcpserver.com/servers/superlocalmemory)

Connect your client

Add in one click

Or add it by command or config

Add in one click

Or add it by command or config

Add in one click

Or add it by command or config

Settings 4

What it does

What it needs

Limits

Who it suits

Tools 58 of 110

Similar servers

agentmemory

SuperSkill

MegaMemory

Engram

projectmem

marm-memory