Smriti

Sanskrit: स्मृति — memory, remembrance A lightning-fast, self-hosted knowledge store and memory layer for AI agents.

---

Why Smriti?

Every AI agent needs memory. Mem0 is cloud-only. Letta is research-heavy. Neither has a knowledge graph.

Smriti is different: self-hosted, knowledge-graph-native, MCP-ready, and fast enough to handle millions of operations. Your data never leaves your machine.

Key Features

• MCP Server — Plug into Claude, GPT, or any MCP-compatible agent instantly
• Knowledge Graph — Notes auto-link via [[wiki-links]]; agents discover connections via graph traversal
• Agent Memory — Key-value store with namespaces, TTL, and tool execution logs
• Full-Text Search — SQLite FTS5 with sub-millisecond queries
• REST API — Full CRUD + graph + agent endpoints on Axum
• Self-Hosted — SQLite database, no cloud dependency, no API costs
• Sync — Cross-device via Synology NAS, WebDAV, or any filesystem mount

How It Compares

Feature	Smriti	Mem0	Letta	LangMem
Self-hosted	Yes	No (cloud)	Yes	Partial
Knowledge graph	Yes	No	No	No
MCP native	Yes	No	No	No
Wiki-links	Yes	No	No	No
Full-text search	FTS5	Vector	Vector	Vector
Language	Rust	Python	Python	Python
TTL support	Yes	No	No	No

---

Quick Start

cargo install smriti

# Create notes with wiki-links — connections are automatic
smriti create "LLM Architecture" \
  --content "Transformers use [[Attention Mechanisms]] for [[Parallel Processing]]"

smriti create "Attention Mechanisms" \
  --content "Self-attention is the core of [[LLM Architecture]]. See also #transformers"

# Search across all notes
smriti search "attention"

# View the knowledge graph
smriti graph

# Start the MCP server (for AI agents)
smriti mcp

# Start the REST API
smriti serve --port 3000

Build from source

git clone https://github.com/smriti-AA/smriti.git
cd smriti
cargo build --release
./target/release/smriti --help

---

MCP Server

Start with smriti mcp. Agents communicate via JSON-RPC 2.0 over stdio.

8 tools available to agents:

Tool	Description
notes_create	Create a note with markdown content. [[wiki-links]] and #tags are auto-detected
notes_read	Read note by ID or title
notes_search	Full-text search across all notes
notes_list	List recent notes, optionally filtered by tag
notes_graph	Get full knowledge graph or subgraph around a note
memory_store	Store key-value memory with optional namespace and TTL
memory_retrieve	Retrieve a memory by agent ID, namespace, and key
memory_list	List all memory entries for an agent

Claude Desktop Integration

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "smriti": {
      "command": "smriti",
      "args": ["mcp", "--db", "/path/to/smriti.db"]
    }
  }
}

Example: Agent Stores and Retrieves Memory

# Agent stores a finding
echo '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"memory_store","arguments":{"agent_id":"researcher-1","key":"finding","value":"Transformers scale logarithmically with data size"}}}' | smriti mcp

# Agent creates a linked note
echo '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"notes_create","arguments":{"title":"Scaling Laws","content":"Key insight: [[Transformer]] performance scales logarithmically. Related to [[Chinchilla]] findings."}}}' | smriti mcp

---

CLI Reference

smriti create <title>       Create a note (--content, --file, --tags)
smriti read <id>            Read a note by ID or title (--json)
smriti list                 List notes (--limit, --tag, --json)
smriti search <query>       Full-text search (--limit)
smriti graph                Knowledge graph (--format json|dot|text, --center)
smriti stats                Database stats + smart link suggestions
smriti serve                REST API server (--host, --port)
smriti mcp                  MCP server over stdio
smriti sync                 Sync with remote (--remote, --direction push|pull|both)
smriti import <dir>         Import .md files (--recursive)
smriti export <dir>         Export to .md files (--frontmatter)

---

REST API

Start with `smriti serve --port 300

{"mcpServers": {"smriti": {"command": "smriti", "args": ["mcp", "--db", "/path/to/smriti.db"]}}}