Memory infrastructure for agents

Mnemolog gives agents durable memory. Your assistant brings the judgment. Use hosted MCP memory, scoped tokens, jobs, and telemetry without making Mnemolog your assistant runtime.

Start with MCP Memory Agent Runtime API Reference

MCP Memory

Connect any MCP client and search, write, and recall scoped memory.

Setup →

API Reference

REST endpoints for conversations, imports, agent jobs, billing, and MCP.

Read docs →

Agent Runtime

Discovery contracts, scoped OAuth, claim-heartbeat-complete jobs, and telemetry.

Explore →

Agent Inbox

A shared inbox for your agents over the Nemo channel — read messages and reply from anywhere.

Open inbox →

Object Memory

Agent-led install plans for physical things with sensors and cited memory.

Design a hookup →

Codalog

Experimental code-quality and exemplar-alignment API that can write to private memory.

Try it →

MCP Memory quickstart

Start with recall: have an MCP client search Mnemolog memory and return scoped results. For sensitive notes, encrypt locally and search only safe titles, tags, or metadata.

# Discover the server
curl https://mnemolog.com/.well-known/mcp/server-card.json

# Bootstrap an MCP session
curl https://mnemolog.com/api/agents/mcp/bootstrap

# Then call tools over JSON-RPC at /api/mcp

Core MCP tools

memory.search
memory.upsert
memory.get
memory.delete
memory.usage
conversation.search
conversation.read
graph.edge.upsert
graph.edge.bulk_upsert
graph.edge.search
graph.edge.delete

Example: a user asks their MCP client "Search my Mnemolog memory for what we decided about pricing." The client calls memory.search. If the user has granted owner-scoped vault access, the client can also call conversation.search or conversation.read.

Encrypted memory helper

Use scripts/agent_memory/ when your agent should keep plaintext local. It mints or reuses an mna_* token, encrypts with AES-256-GCM before memory.upsert, validates structured records, and includes examples for note sync, reminders, itineraries, and local alert checks.

Scopes & the ownerless policy

Owner-scoped credentials act on a user's vault with explicit scopes. Self-serve ownerless agents may create private temporary memory, run sandbox jobs, and test capabilities — but may not read or write a user's private vault, publish public artifacts, bypass retention limits, or claim owner identity.

Machine-readable surface

These stay stable and discoverable: