The agent-native database platform on Base. Real Postgres. Real schemas. Wallet-auth. MCP-native. Memory in gold.
🟢 Live MVP — try it at mnemedb.dev or smoke-test the
gateway: gateway.mnemedb.dev/health.
Mneme is a managed multi-tenant Postgres platform built for AI agents. Each agent gets a dedicated schema with four memory tables ready to go, then can create as many custom tables as it needs at runtime — text, int, jsonb, vector, uuid, anything Postgres + pgvector supports. No API keys, no DB credentials, no infrastructure: agents authenticate with their wallet, and everything is one SDK call (or one MCP tool call) away.
agent ─wallet-signed req─▶ mneme-gateway ─lookup project─▶ agent_<handle>.* on OUR Postgres
Mem0 / Pinecone / Letta give you a fixed memory store. Mneme gives you a Postgres schema you control: create tables, drop tables, run vector searches over any column, all over the same wallet identity. The four default tables are a head start, not a ceiling.
| Feature | Mneme | Mem0 / Pinecone |
|---|---|---|
| Custom tables at runtime | ✅ via mneme.createTable(…) |
❌ |
| Vector search on any column | ✅ | partial |
| jsonb / text / int / uuid / date | ✅ | ❌ |
| Per-wallet schema isolation | ✅ | per-API-key |
| Wallet auth (no API keys) | ✅ | ❌ |
| MCP server for agents | ✅ | growing |
| Onchain identity (Phase 2) | ✅ | ❌ |
- Our Postgres cluster (Neon for MVP). Every Mneme project = one schema, fully isolated.
- Wallet auth. EIP-712 typed-data on every request (or 24-hour session JWT). Smart wallets (Coinbase Smart Wallet, Safe) via ERC-1271/6492.
- Default tables.
memories,documents,events,kvsprovisioned at project create. - Runtime DDL.
POST /v1/tablesto create any table; columns: text/int/bigint/bool/jsonb/timestamptz/vector/numeric/uuid/date. - Generic vector search. KNN over any
vector(N)column on any table. - MCP server. Five tools —
mneme_create_table,mneme_list_tables,mneme_insert,mneme_list,mneme_vector_search— for Claude / Cursor / Cline. - Dashboard. Connect wallet, pick handle, see your schema + tables + stats + MCP config.
| Workspace | Description |
|---|---|
contracts/ |
Foundry — AgentRegistry (Phase 2 onchain handle layer) |
gateway/ |
Bun + Hono — direct Postgres, schema-per-project, runtime DDL, wallet auth |
sdk/ |
mneme-sdk — m.createTable, m.from(table).insert / list, m.vectorSearch, … |
mcp/ |
mneme-mcp — Claude/Cursor MCP tools |
dashboard/ |
Vite + React + TS + Tailwind + wagmi — pick handle, see tables, MCP config |
scripts/smoke.ts |
End-to-end smoke test (create project, custom table, insert, list, vector search) |
import { privateKeyToAccount } from "viem/accounts";
import { Mneme } from "mneme-sdk";
const m = new Mneme({
account: privateKeyToAccount(process.env.AGENT_PRIVATE_KEY as `0x${string}`),
gatewayUrl: "https://gateway.mnemedb.dev",
});
// Use the defaults
await m.memories.insert({ text: "hello", embedding: vec });
// Or create your own tables
await m.createTable({
name: "users",
columns: [
{ name: "wallet", type: "text", unique: true, nullable: false },
{ name: "twitter", type: "text" },
{ name: "embedding", type: "vector", dim: 1536 },
],
});
await m.from("users").insert({ wallet: "0xabc…", twitter: "alice", embedding: vec });
// Vector search across any vector column
const { matches } = await m.vectorSearch({
table: "users", column: "embedding", embedding: query, k: 5,
});- Visit dashboard → connect wallet (Coinbase Smart Wallet, MetaMask, …).
- Pick a handle (
alice). - Wallet signs
CreateProject→ schemaagent_alice+ 4 default tables provisioned in one transaction. - SDK / MCP / dashboard all work against
agent_aliceimmediately. - Agents call
mneme.createTable(…)as they need new shapes.
- Postgres with pgvector — Neon free tier works for MVP.
gateway/.env:DATABASE_URL,GATEWAY_JWT_SECRET(32-byte hex).- Deploy gateway (Fly / Railway / Render). Deploy dashboard (Vercel).
$MNEME — Day-1 launch with public MVP, Clanker / Flaunch on Base. Utility: gateway query metering + stake-for-discount + (Phase 2) onchain permissions registry stake.
Pre-MVP. Stealth. Smoke test passes end-to-end (project create → custom table → insert → list → vector search → stats).