Memory for Agents
Persistent context that follows your AI across sessions. Automatic ingestion, confidence-scored facts, zero commands needed.
Hosted version coming soon — hello@getplumb.dev
AI agents have no long-term memory
Every session starts from scratch
Your agent forgets what it learned yesterday. You repeat yourself, every time.
"Remember this" shouldn't be a command
Manual memory management breaks flow. Context should be captured automatically.
Context is scattered across tools
Notes in one place, chat history in another. No single source of truth for what your agent knows.
How it works
Ingest
Agents talk. Plumb listens. Every conversation is logged automatically via the MCP protocol — no manual saves, no commands.
User: I prefer spaces over tabs Agent: Noted, I'll use spaces for indentation
Extract
Facts are pulled from raw conversation logs, deduplicated, and scored by confidence. Stale facts decay over time so your agent stays current.
{
"subject": "user",
"predicate": "prefers",
"object": "spaces over tabs",
"confidence": 0.95,
"decay": 0.98
}Retrieve
When your agent needs context, a single MCP tool call returns the most relevant facts. Drop this into your config and it just works.
{
"mcpServers": {
"plumb": {
"command": "plumb",
"args": ["serve", "--mcp"],
"env": {
"PLUMB_STORE": "~/.plumb/store.db"
}
}
}
}Simple pricing
Start free with the open-source core. Upgrade when you want hosted infrastructure.
OSS
Self-host on your own machine. MIT licensed.
- SQLite local storage
- Automatic conversation ingestion
- Confidence-scored fact extraction
- Recency decay
- MCP server included
- Unlimited sessions
Solo
Hosted infrastructure. Zero ops.
- Everything in OSS
- Postgres + pgvector on Supabase
- Cross-device sync
- Semantic search over facts
- Automatic backups
- Priority support