IronMem
BMC-INC/Iron-mem
Memória persistente de sessão para assistentes de IA em desenvolvimento
Ver no GitHubPlugin que mantém memória persistente para assistentes de IA em desenvolvimento, registrando sessões de codificação e injetando contexto automaticamente nas próximas. Funciona com Claude Code, Cursor, Copilot e Windsurf como binário único em Rust, local-first, eliminando a necessidade de re-explicar projetos.
Instalação
npx convez-ai BMC-INC/Iron-mem1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Manter contexto de projeto entre sessões de codificação
- Recuperar automáticamente arquitetura e decisões técnicas anteriores
- Integrar com múltiplos IDEs e assistentes (Claude, Cursor, Copilot, Windsurf)
Veredito

Solução robusta em Rust que resolve um problema real: context-bleeding entre sessões de IA. Implementação sofisticada com tipagem de memórias, compressão reversível e suporte multi-provider, mas requer setup técnico e conhecimento de MCP servers para máximo proveito.
README
A tradução para o português deste README ainda está na fila.
What's New in v0.4.0
Lossless, reversible memory — plus a real memory model: scoping, typed memories, an always-on user profile, and a correction miner.
- CCR — losslessly reversible memory (Headroom pattern) — every truncated tool output and the verbatim pre-LLM session transcript is preserved in a content-addressed, deduplicated, byte-exact compressed blob store inside the DB. The new
retrieve_originaltool pulls the exact original back behind any compressed memory. No more lossy truncation. - Memory scoping & typed memories (Supermemory patterns) — memories carry a scope (
projectvs.user/cross-project) and a kind (session,error_solution,preference,architecture,learned_pattern,project_config,profile). Session-start injection ranks project ∪ user memories and boosts durable kinds. remembertool — store an explicit, typed memory in one call (scope,kind,text,tags). User-scope facts follow you into every project.- User profile — your cross-project memories are distilled into a single always-injected profile (LLM summary, or a deterministic local rollup when offline). Read/regenerate with
get_profile/refresh_profile. - Correction miner — error→fix loops in a session (a failing command, edits, then the same command passing) are mined into
error_solutionmemories and surfaced vialist_corrections, so past fixes resurface when the work recurs. - 18 MCP tools now — adds
retrieve_original,remember,get_profile,refresh_profile,list_corrections. - Still zero telemetry. Still local-first. Your data stays yours.
- Multi-provider compression — use OpenAI, Google Gemini, or Anthropic as your LLM. Set
"provider": "openai"in settings. - Neovim plugin — native Lua plugin with auto session lifecycle,
:IronMemSearch,:IronMemStatus - Windows support —
install.ps1, platform-aware messages, robust home directory detection - Web UI — browse, search, and delete memories at
http://localhost:37778/ui - Discovery tools — list known projects, search across all projects, and inspect per-project session history
- Still zero telemetry. Still local-first. Your data stays yours.
- 13 MCP tools — session_start, session_end, record_event, compress_session, get_context, get_status, list_memories, search_memories, search_global, list_projects, list_sessions, inject_context, wipe_project
- Dual database — SQLite (local, FTS5 full-text search) + Postgres (self-hosted, tsvector) via
DATABASE_URL - Every MCP client — Claude Desktop, Claude Code, Cursor, Windsurf, ChatGPT Desktop, Zed, and more
- Docker deployment —
docker-compose upfor remote/team setups with Postgres ironmem mcp— new subcommand for direct MCP stdio transport (Claude Desktop/Code)- REST server still works — existing hooks and curl-based workflows unaffected
IronMem gives AI coding tools persistent memory across sessions. It silently records what happened during your session, compresses it into concise memory, and injects that context into your next session automatically.
No copy-pasting. No rebuilding context from scratch. No "remember when we refactored auth yesterday?"
Works with every major AI coding tool — Claude Code, Claude Desktop, Cursor, Windsurf, ChatGPT Desktop, GitHub Copilot, Zed, VS Code, and any MCP-compatible client.
Compress with the LLM you already pay for — Anthropic Claude, OpenAI GPT-4o, or Google Gemini. Switch providers with one config change.
Free and open source. Runs locally or on your own infrastructure. No telemetry. No cloud dependency. No subscription. SQLite or Postgres storage. Plain markdown output. Single Rust binary.
Why this exists
AI coding tools are great inside a session and terrible across sessions. They help you ship faster, but every fresh session forgets your architecture decisions, debugging trail, and what changed yesterday.
IronMem fixes the handoff.
Before vs after
Without IronMem:
"We already changed the auth middleware, switched to JWT, updated the migration, and fixed the failing tests in billing. Let me explain the whole thing again."
With IronMem:
Open a new session. Your assistant already has the recent project context.
Quick Start
- Install IronMem:
curl -fsSL https://raw.githubusercontent.com/BMC-INC/Iron-mem/main/install.sh | bash - Add your API key to IronMem's key file:
echo "your-key-here" > ~/.ironmem/api_key && chmod 600 ~/.ironmem/api_keyPrefer the key file over
export ANTHROPIC_API_KEY. Claude Code (and some other tools) bill againstANTHROPIC_API_KEYwhenever it's set in your shell — using pay-as-you-go API credit instead of your Claude subscription. The key file keeps IronMem's key out of your environment so it can't change how other tools bill. (IronMem still honorsANTHROPIC_API_KEYif you prefer the env var.) - Start coding! IronMem handles the rest silently in the background.
Table of Contents
- Quick Start
- The Problem
- The Fix
- Who Should Use This?
- How It Works
- Install
- CLI
- Multi-Provider Support
- MCP Setup
- Web UI
- Configuration
- Troubleshooting
- Architecture
- Why Rust?
- Design Principles
- Why not just use CLAUDE.md?
- Roadmap
- Contributing
- Support
- License
The Problem
Every time you start a new session with Claude Code, Cursor, Copilot, or any AI coding assistant — it starts from zero. It doesn't know what you built yesterday. It doesn't know what broke. It doesn't know what you decided.
You end up re-explaining context every single session.
The Fix
IronMem silently records what happens during your coding session, compresses it into a concise memory using your LLM provider of choice (Anthropic, OpenAI, or Gemini), and injects that context into your next session automatically.
No setup per session. No copy-pasting. No "remember when we..."
Without IronMem: "Hey Claude, remember yesterday we refactored the auth middleware and switched to JWT? And the database migration for the users table? And..."
With IronMem: You open a new session. It already knows.
Who Should Use This?
IronMem is designed for:
- Developers frustrated with re-explaining context to AI tools every single session.
- Teams working on large, multi-session projects where context gets easily lost.
- Developers frequently switching between multiple AI tools like Copilot, Claude Code, Windsurf, or Cursor.
- Solo developers who want to maintain flow and continuity without manual effort.
How It Works
flowchart LR
A["🟢 Session Start"] -->|inject memories| B["📄 IRONMEM.md"]
B --> C["🤖 AI reads context"]
C --> D["🔧 You code"]
D -->|every tool call| E["🗄️ SQLite"]
D --> F["🔴 Session End"]
F -->|compress via LLM| G["🧠 Memory"]
G -->|next session| A
style A fill:#22c55e,color:#fff,stroke:none
style F fill:#ef4444,color:#fff,stroke:none
style G fill:#8b5cf6,color:#fff,stroke:none
style E fill:#0ea5e9,color:#fff,stroke:none
Everything runs locally. Your data stays on your machine.
Install
curl -fsSL https://raw.githubusercontent.com/BMC-INC/Iron-mem/main/install.sh | bash
Or clone and build manually:
git clone https://github.com/BMC-INC/Iron-mem.git
cd Iron-mem
chmod +x install.sh
./install.sh
Windows:
git clone https://github.com/BMC-INC/Iron-mem.git
cd Iron-mem
powershell -ExecutionPolicy Bypass -File install.ps1
Add IronMem to your PATH (in ~/.zshrc or ~/.bashrc) and write your API key to IronMem's key file:
export PATH="$HOME/.ironmem/bin:$PATH" # in ~/.zshrc / ~/.bashrc
echo "your-key-here" > ~/.ironmem/api_key && chmod 600 ~/.ironmem/api_key
Use the key file, not
export ANTHROPIC_API_KEY— a globalANTHROPIC_API_KEYmakes Claude Code (and similar tools) bill against pay-as-you-go API credit instead of your subscription. IronMem reads~/.ironmem/api_keyautomatically.
Restart your terminal and Claude Code. That's it.
Requirements: Rust/Cargo (the installer will tell you if it's missing)
CLI
ironmem server # Start REST + MCP SSE server
ironmem mcp # Start MCP stdio server (for Claude Desktop/Code)
ironmem serve # Start SSE server with bearer token auth
ironmem serve --public # Same + Cloudflare Tunnel for remote MCP clients
ironmem serve --public --no-auth # Authless public tunnel for claude.ai personal use
ironmem status # Health check + DB stats
ironmem projects # All projects with stored memories
ironmem list # Recent memories for current project
ironmem search "auth middleware" # Hybrid (keyword + semantic) search across memories
ironmem search-global "auth middleware" # Search across all projects
ironmem sessions # Session history for current project
ironmem inject # Manually rebuild IRONMEM.md (relevance-ranked)
ironmem remember "..." # Store an explicit memory (--scope user, --kind preference, --tags)
ironmem profile # Show the user profile (--refresh to regenerate it)
ironmem corrections # List mined error→fix memories (--all for every project)
ironmem compress <id> # Manually compress a session
ironmem embed # Backfill semantic embeddings for existing memories
ironmem gc # Reclaim unreferenced CCR blobs (after wipes)
ironmem wipe # Delete all memories for current project
ironmem config # Print current settings
Multi-Provider Support
IronMem works as an MCP server (native integration) or via IRONMEM.md (plain markdown, universal):
| Platform | MCP Native | IRONMEM.md | Setup |
|---|---|---|---|
| Claude Code | Yes | Yes | Setup → |
| Claude Desktop | Yes | Yes | Setup → |
| claude.ai | Yes | Yes | Setup → |
| Cursor | Yes | Yes | Setup → |
| Windsurf | Yes | Yes | Setup → |
| ChatGPT Desktop | Yes | — | Setup → |
| Zed | Yes | — | Setup → |
| VS Code (Copilot/Continue/Cline) | Yes | Yes | Setup → |
| Any MCP Client | Yes | — | stdio or SSE transport |
| Any AI Tool | — | Yes | Read IRONMEM.md as project context |
MCP Setup
IronMem supports two MCP transports:
- stdio — for local clients that launch the server themselves (Claude Code, Claude Desktop, Cursor)
- Streamable HTTP — for remote/cloud clients that connect over HTTP. Uses standard request/response and bearer-token auth, so it works through tunnels and reverse proxies for clients that support static bearer tokens.
Once connected over MCP, IronMem now supports project discovery directly:
list_projects— show every project with stored memoriessearch_global— search across all projectslist_sessions— inspect session history inside a project
Claude Code MCP Setup
Claude Code connects via stdio — it launches ironmem mcp directly.
Option A: CLI (recommended)
claude mcp add ironmem -- ~/.ironmem/bin/ironmem mcp
Option B: Project .mcp.json (share with your team)
Create .mcp.json in your project root:
{
"mcpServers": {
"ironmem": {
"command": "~/.ironmem/bin/ironmem",
"args": ["mcp"],
"env": {
"ANTHROPIC_API_KEY": "your-key-here"
}
}
}
}
Note: Claude Code hooks (installed by
install.sh) and MCP can coexist. The hooks use the REST API for automatic observation recording; MCP gives you direct tool access. You can use both, or just one.
Claude Desktop MCP Setup
Claude Desktop also connects via stdio.
Add to your claude_desktop_config.json:
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"ironmem": {
"command": "/Users/YOU/.ironmem/bin/ironmem",
"args": ["mcp"],
"env": {
"ANTHROPIC_API_KEY": "your-key-here"
}
}
}
}
Replace /Users/YOU with your actual home directory path. Restart Claude Desktop after saving.
claude.ai (Web)
claude.ai runs in the cloud, so it cannot reach localhost.
IronMem is a local-first tool. The recommended setup for full MCP access is Claude Code or Claude Desktop using stdio.
Anthropic's current claude.ai custom connector UI supports authless and OAuth-based remote MCP servers, but not a manual static bearer-token field. For personal use, the honest compatibility path is an authless ephemeral tunnel:
ironmem serve --public --no-auth
That command does three things:
- Starts the SSE server with no auth
- Launches a Cloudflare Tunnel (free, no account needed) to expose it publicly
- Prints the public URL
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
IronMem SSE Server
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Local: http://127.0.0.1:37779/mcp
Auth: Disabled (--no-auth)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Public URL: https://xxx-yyy-zzz.trycloudflare.com
Remote MCP setup:
URL: https://xxx-yyy-zzz.trycloudflare.com/mcp
Auth: None
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
In claude.ai:
- Open
Settings - Open
Integrations - Choose
Add custom connector - Set
NametoIronMem - Paste the printed
https://...trycloudflare.com/mcpURL intoRemote MCP server URL - Leave the OAuth fields blank
The trycloudflare.com URL is ephemeral and changes whenever you restart the public tunnel, so update your connector URL each time you relaunch ironmem serve --public --no-auth.
This is no longer local-only. The tunnel exposes your MCP endpoint over the internet for as long as it is running.
For a personal local tool, this tradeoff is often acceptable because the URL is short-lived and changes on each restart. Still, use --no-auth deliberately and only when you understand that you are trading security for compatibility.
Without --no-auth: ironmem serve and ironmem serve --public use bearer-token auth for clients that support static bearer tokens.
Requirements: Install cloudflared for best results (brew install cloudflared on macOS, winget install Cloudflare.cloudflared on Windows). Falls back to npx cloudflared if not installed.
Cursor / Windsurf MCP Setup
Both use stdio. Add to your MCP settings:
Cursor: Settings → MCP → Add Server
Windsurf: Settings → MCP → Add Server
{
"ironmem": {
"command": "~/.ironmem/bin/ironmem",
"args": ["mcp"]
}
}
Other MCP Clients
Any MCP client that supports stdio transport can use IronMem:
{
"command": "~/.ironmem/bin/ironmem",
"args": ["mcp"]
}
For clients that support Streamable HTTP, start the server and point the client at http://localhost:37779/mcp:
ironmem serve
Neovim Plugin
IronMem includes a native Neovim plugin that communicates via MCP stdio.
Install with lazy.nvim:
{
"BMC-INC/Iron-mem",
config = function()
require("ironmem").setup({
-- binary = "~/.ironmem/bin/ironmem", -- default
-- auto_start = true, -- session_start on VimEnter
-- auto_end = true, -- session_end on VimLeavePre
-- record_events = true, -- record buffer writes
})
end,
}
Commands:
| Command | Description |
|---|---|
:IronMemStart | Manually start a session |
:IronMemEnd | End session and compress |
:IronMemStatus | Show database stats |
:IronMemSearch <query> | Search memories in a split buffer |
Web UI
When the REST server is running, a built-in memory browser is available at:
http://localhost:37778/ui
The UI shows sessions, memories, and database stats. You can browse, search, and delete memories directly from the browser.
Configuration
~/.ironmem/settings.json:
{
"port": 37778,
"provider": "anthropic",
"model": "claude-sonnet-4-6-20250627",
"inject_limit": 5,
"max_observation_bytes": 2048,
"db_path": "/Users/you/.ironmem/mem.db",
"database_url": null,
"mcp_transport": "stdio",
"mcp_sse_port": 37779,
"auth_token": null,
"embedding": {
"provider": "auto",
"model": null,
"ollama_url": "http://localhost:11434",
"weights": { "relevance": 0.5, "recency": 0.3, "importance": 0.2 },
"recency_half_life_days": 30
}
}
All fields optional. Sensible defaults provided. auth_token is generated automatically the first time you run ironmem serve without --no-auth. The embedding block is optional — omit it entirely and IronMem behaves exactly as before (keyword-only search, recency injection).
Semantic Search & Embeddings
IronMem can blend keyword (FTS) and semantic (vector) retrieval using Reciprocal Rank Fusion, and rank session-start injection by a blend of relevance + recency + importance. Embeddings are stored locally in SQLite via sqlite-vec (or pgvector on Postgres); nothing is sent anywhere unless you explicitly choose an API provider.
Privacy posture: this is a governance tool, so the default is local-first / no data egress. The auto provider prefers a local embedder and silently degrades to keyword-only search if none is available — it never phones home and never hard-fails a command.
embedding.provider | Behavior | Data egress |
|---|---|---|
"auto" (default) | Use local Ollama if reachable, else the built-in ONNX model (if compiled in), else keyword-only | None (unless only an API key is configured) |
"ollama" | Local Ollama embeddings | None (localhost) |
"onnx" | In-process ONNX model (requires --features local-onnx build) | None |
"openai" | OpenAI embeddings API | Sends memory text to OpenAI |
"google" | Google embeddings API | Sends memory text to Google |
"none" | Keyword-only FTS + recency injection (legacy behavior) | None |
Recommended local setup (no egress):
# Install Ollama, then pull an embedding model:
ollama pull nomic-embed-text
# IronMem's "auto" provider will detect and use it automatically.
Built-in ONNX (no Ollama, no network): compile with the optional feature so embeddings run fully in-process:
cargo install --path . --features local-onnx
Blend weights (embedding.weights) control session-start injection ranking: relevance (semantic match to your current git context), recency (true half-life decay set by recency_half_life_days), and importance (an LLM-assigned 1–10 score per memory). They need not sum to 1.
Backfill existing memories — after enabling embeddings, index memories created before:
ironmem embed # embed memories missing a vector (all projects)
ironmem embed --project . # scope to one project
ironmem embed --force # rebuild the whole index from scratch
Provider
IronMem supports three LLM providers for session compression:
| Provider | provider value | Default model | API key env var |
|---|---|---|---|
| Anthropic | "anthropic" | claude-sonnet-4-6-20250627 | ANTHROPIC_API_KEY |
| OpenAI | "openai" | gpt-4o | OPENAI_API_KEY |
| Google Gemini | "google" | gemini-2.0-flash | GOOGLE_API_KEY |
To switch providers, set "provider" in ~/.ironmem/settings.json and ensure the corresponding API key is set. The model field overrides the provider's default model.
Environment Variables
| Variable | Default | Description |
|---|---|---|
DATABASE_URL | (none) | Postgres URL. Overrides db_path when set. |
IRONMEM_MCP_TRANSPORT | stdio | MCP transport: stdio or sse |
ANTHROPIC_API_KEY | (none) | Required when provider is anthropic (default) |
OPENAI_API_KEY | (none) | Required when provider is openai |
GOOGLE_API_KEY | (none) | Required when provider is google |
API Key
IronMem needs an LLM API key to compress session observations into memories.
Recommended (Anthropic): write the key to ~/.ironmem/api_key (echo "your-key" > ~/.ironmem/api_key && chmod 600 ~/.ironmem/api_key). Keeping it in this file rather than a global ANTHROPIC_API_KEY export avoids changing how other tools bill — Claude Code, for instance, charges pay-as-you-go API credit whenever ANTHROPIC_API_KEY is set in the environment, instead of using your subscription. IronMem reads the file automatically (it also works when the background server is spawned via nohup, where some shells strip env vars from child processes).
IronMem still honors the per-provider environment variable if you prefer it — ANTHROPIC_API_KEY, OPENAI_API_KEY, or GOOGLE_API_KEY. The file fallback applies to the default Anthropic provider.
Troubleshooting
Server not starting:
ironmem status # Check if server responds
cat ~/.ironmem/server.log # Check server logs
~/.ironmem/bin/ironmem server # Run manually to see errors
Observations not being recorded:
ironmem status # Check observation count
sqlite3 ~/.ironmem/mem.db "SELECT count(*) FROM observations;"
If count stays at 0, your hooks may not be installed. Re-run ./install.sh or check that ~/.claude/hooks/post-tool-use.sh exists and is executable.
Compression failing (memories always 0):
# Check if the API key is accessible (the key file is the recommended source)
cat ~/.ironmem/api_key # Should contain your key
echo $ANTHROPIC_API_KEY # Optional env-var fallback (may be empty)
# Try manual compression
ironmem compress <session-id> # Get session ID from server.log
Hooks not firing:
Check that ~/.claude/settings.json has the hooks registered under the "hooks" key. Re-running ./install.sh will fix this.
Architecture
~/.ironmem/
├── bin/ironmem # Single compiled binary
├── mem.db # SQLite DB: memories (FTS5) + vectors + CCR blob store
├── settings.json # Configuration
├── api_key # Anthropic API key (chmod 600; keeps it out of your shell env)
├── current_session # Active session ID (ephemeral)
└── server.log # Worker logs
~/.claude/hooks/ # Auto-installed Claude Code hooks
├── session-start.sh # Injects memories on session start
├── post-tool-use.sh # Records every tool call
├── stop.sh # Triggers compression
└── session-end.sh # Cleanup
~9,000 lines of Rust. MCP-native. SQLite or Postgres. Lossless, reversible memory. One binary. No external runtimes.
Why Rust?
Rust was chosen for IronMem to deliver:
- Maximum Performance: Minimal overhead and lightning-fast execution, essential for a tool that hooks into every single CLI command.
- Zero Dependencies: Compiles down to a single binary. No need to install Python, Node.js, or complex runtime environments.
- Memory Safety & Reliability: Guaranteed safety without a garbage collector ensures the background worker remains rock-solid and leak-free.
Design Principles
- Zero friction — hooks run silently, never interrupt your workflow
- Local-first — runs on your machine by default, your data stays yours
- MCP-native — speaks the protocol every major AI client is adopting
- Provider-agnostic — MCP for native integration, plain markdown for everything else
- Self-hostable — Docker + Postgres for team deployments, still zero cloud dependencies
- Fail-safe — if IronMem crashes, your coding session is unaffected
Who this is for
IronMem is for developers who use AI coding tools heavily and want continuity across sessions.
It is especially useful if you:
- switch between Claude Code, Cursor, Copilot, or Windsurf
- work on projects that span many sessions
- are tired of re-explaining architecture, bugs, and recent changes
- want local-first memory instead of a hosted service
Who this is not for
IronMem is not trying to be:
- a generic memory platform for every kind of agent
- a cloud sync product
- a team knowledge base
- a dashboard-heavy workflow tool
It solves one narrow problem well: session memory for AI coding workflows.
Why not just use CLAUDE.md?
CLAUDE.md is great for static project context — things like "use tabs not spaces" or "we use Axum for routing." But it's manual. You write it, you maintain it, and it doesn't know what happened last session.
IronMem is automatic and session-aware:
| CLAUDE.md | IronMem | |
|---|---|---|
| Updates | You write it manually | Auto-generated from session activity |
| Scope | Static project rules | Dynamic session history |
| Rotation | You manage it | Old memories age out automatically |
| Search | Ctrl+F | Full-text search across all sessions |
| Effort | High | Zero — hooks handle everything |
They work together. CLAUDE.md holds your project rules. IronMem holds what happened.
Docker Deployment
Run IronMem with Postgres for team/remote setups:
ANTHROPIC_API_KEY=your-key docker-compose up --build
This starts IronMem with Streamable HTTP on http://localhost:37779/mcp and Postgres 16, plus the REST server on http://localhost:37778.
Roadmap
Shipped in v0.2.0
- MCP-native server (stdio + Streamable HTTP)
- Dual database — SQLite (local, FTS5) + Postgres (self-hosted)
- Docker deployment with Postgres
- Bearer token authentication
-
ironmem serve --publicwith Cloudflare Tunnel for remote MCP clients -
ironmem serve --public --no-authfor claude.ai personal use - Works with Claude Code, Claude Desktop, Cursor, Windsurf, ChatGPT Desktop, Zed, VS Code
Shipped in v0.3.0
- Multi-provider compression — OpenAI, Google Gemini, or Anthropic (configurable via
providerin settings) - Neovim plugin (
nvim/lua/ironmem/) — auto session lifecycle,:IronMemSearch,:IronMemStatus - Windows native support —
install.ps1, platform-aware install messages, robust home dir detection - Web UI memory browser —
http://localhost:37778/uiwhen REST server is running
Shipped in v0.4.0
- Semantic foundation — hybrid FTS + vector (RRF) retrieval, local-first embeddings, and relevance-ranked session-start injection
- Reliability & security hardening — stdio MCP stream is no longer corrupted by log output, UTF-8-safe truncation prevents a crash on multibyte tool output, and 7 dependency advisories were patched
- CCR — losslessly reversible memory (Headroom pattern): a content-addressed, deduplicated blob store + byte-exact per-content-type compression (zstd + per-type dictionaries) + a
retrieve_originaltool, so the verbatim original behind any compressed memory is always recoverable — no more lossy truncation. Refcount GC (ironmem gc) + storage stats inget_status. Design » - Memory scoping & types (Supermemory patterns): project vs. user (cross-project) scope, typed memories (
error_solution/preference/architecture/learned_pattern/ …), scope-aware injection with per-kind boosts, and aremembertool - Always-injected user profile — cross-project facts distilled into one profile memory (LLM summary or deterministic local rollup);
get_profile/refresh_profile - Correction miner — error→fix loops become
error_solutionmemories, surfaced vialist_corrections
Next
- Bespoke per-content-type transforms — invertible log timestamp-delta / diff-token / AST-aware code normalization on top of the dictionary codecs (currently documented-future; the byte-exact contract is the gate)
- Observation-blob lifecycle GC — reclaim CCR blobs behind deleted observations (memory-session blobs are already GC'd)
Contributing
Contributions are welcome. Please read CONTRIBUTING.md before opening a PR.
TL;DR: Open an issue first. Bug fixes and provider compatibility improvements are always welcome. We don't accept changes that add runtime dependencies or complexity.
⭐ Support
If you find IronMem useful, please consider giving it a star! 🌟 This helps others discover the project and motivates further development. Contributions, issues, and feature requests are also highly welcome.
License
Apache-2.0 © 2026 ExecLayer Inc
Maintainer: James Benton