Nenavi Cloud
Nenavi Cloud is the hosted service built on PRME. You send it text over HTTP or MCP, and it keeps long-term memory for your agent and returns the part that matters for a question. You do not run a database, an embedding model, or an extraction model.
Early access. Nenavi Cloud is an alpha for invited users, and the waitlist is the way in. These pages describe how the alpha works today. There is no billing and there is no service level agreement. Limits, routes, and behavior can change. Join the waitlist on the Nenavi site.
PRME is the open-source engine that Nenavi Cloud runs. If you want to run the engine yourself, in your own process or on your own PostgreSQL, use the PRME docs instead. This section covers only the hosted service.
How it differs from PRME
| PRME (open source) | Nenavi Cloud (hosted) | |
|---|---|---|
| Where it runs | In your process, or on a server you run | On Nenavi’s servers |
| How you call it | The Python MemoryClient, the CLI, or an HTTP server you start | HTTP requests and an MCP endpoint, with an API key |
| Where memory is kept | A local pack, or your PostgreSQL | A separate, private memory space for each project |
| Extraction | Your own model provider and your own key | Run by Nenavi, which pays for it. Monthly caps apply |
| Limits | Your hardware | Per-project limits for requests, ingest calls, content tokens, and memories |
| Routes | Every route in the HTTP API reference | The same routes, with a few changes listed in The API |
The Python MemoryClient runs the engine on your machine. It does not call Nenavi Cloud. To use the hosted service from Python, send plain HTTP requests, as the quick start shows.
The parts
- Organization. The account that owns your projects and the people who use them. An organization has owners and members.
- Project. A separate memory space. A project has its own API keys, its own usage, and its own limits. One project’s key can never read another project’s memory.
- API key. A secret that lets a program use one project. You create keys in the console and see the full key once.
user_id. A name you choose for one of your own end users. Every store and retrieve names one. The service does not check it against anything, so your code must send the right one.
What to read next
- Quick start: create a project and a key, then store and retrieve a memory.
- The API: the base URL, authentication, errors, and how the hosted routes differ from PRME.
- Limits and errors: the monthly caps, the rate limit, and every error code.
- MCP: connect Claude Code, Cursor, or Claude Desktop.