1
Arm your IDE
Tell your assistant (Cursor, Claude, …) to use the boilerplate,
@mnemosyne_os/sdk, the MCP server and the
llms.txt from the repo. Have it clone a reference MIT cartridge
and read its src/. It then codes against the real API, not
against its own memory.
The starting prompt:
Prompt Build a Mnemosyne OS cartridge: [describe your app]. Read thellms.txtat the root of github.com/Mnemosyne-OS/Mnemosyne-Neural-OS, clone the MIT cartridgeMnemoReaderand read itssrc/. Connect to the vault with@mnemosyne_os/sdk(and the MCP if needed), under FGAC. Follow the reference patterns, don't guess the API.
2
Find the idea, and spot what goes into memory
Ask one question before writing any code. What does your app need to remember? Notes, decisions, history, entities. That is what you ingest. Everything else is ordinary UI.
3
Ask, and build
Describe the feature, and the assistant codes against the SDK.
ingest() writes into the vault. query() and
ask() read it by meaning, under FGAC and at limited
scope. You iterate exactly like on any other app.
4
Memory is the back-end
No database to provision, no API to host, no auth to wire. The local vault is your persistence, already there, already governed by the human. You get the whole back-end for free. You never manage anyone's data for them.
flowchart TB
A["Your idea"] --> B{"What needs to be in memory?"}
B -->|worth keeping| C["SDK · ingest()"]
B -->|everything else| D["Regular React UI"]
C --> V[("Local vault · FGAC")]
D --> E["SDK · query() / ask()"]
E --> V
V --> F["Answer by meaning"]
F --> D Ready?
Download, request your free license, and get started.
The OS is free for personal use. The founder license is free for devs who build, with $5 of cloud credits to test without an API key. The developers page carries the full reference for the SDK, the sandbox covenant and the open-core.
Not up for coding? We build it for you →