smriti

Local-first persistent memory for AI agents via MCP.

*Smriti (स्मृति) — Sanskrit for "memory, remembrance"*

One brain. Every agent. Zero cloud. Zero cost.

A standalone MCP server backed by sqlite-vec and a local embedding model. Install it, point any MCP-compatible agent at it, and every AI you use shares one persistent, semantically searchable memory.

| | Smriti | Cloud alternatives | |---|---|---| | Setup | npx smriti | Accounts + API keys + config | | Cost | $0 | Variable | | Privacy | 100% local | Data on external servers | | Offline | Full functionality | Needs internet | | Portability | Single .db file | DB export/migration |

npm install -g smriti

# MCP server — stdio mode (for Claude Code, Cursor, etc.)
smriti

# MCP server — HTTP mode (for remote agents)
smriti --http --port 3838

# Capture a memory
smriti capture "We chose PostgreSQL over MySQL for the new service"

# Semantic search
smriti search "database decisions"

# Browse recent memories (last 7 days)
smriti recall --days 7

# Browse by topic
smriti recall "authentication"

# Batch-extract memories from a conversation or document
smriti ingest --text "Long meeting notes or conversation log..." --threshold 0.4

# Memory stats
smriti stats

# Sync to GitHub (requires: smriti auth)
smriti sync

# GitHub auth
smriti auth
smriti whoami
smriti logout

Add to ~/.claude/mcp.json:

{
  "mcpServers": {
    "memory": {
      "command": "smriti"
    }
  }
}

Add to MCP settings:

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["smriti"]
    }
  }
}

smriti --http --port 3838

Then point your client at http://localhost:3838/mcp.

| Tool | Description | |------|-------------| | capture | Store a thought