MythSheetIn development
AI & BYOK
How MythSheet plans to use your own model key, what is experimental today, and what data leaves your machine.
Experimental
AI in MythSheet is experimental. Full end-to-end behaviour with real models has not been verified. Do not rely on it for real work yet.
The direction
MythSheet is bring-your-own-key: you connect a model API, and the AI works through tools to read and analyse data and propose changes. Changes apply only after you approve them.
What exists today
- Experimental inputs for a provider, endpoint, model and key.
- A preview of the AI's proposed changes.
- A code path in the desktop app that applies a proposal after you approve it.
Which providers and models are recommended is coming soon to these docs.
Data flow
Local first does not mean nothing ever leaves your machine:
- You configure a provider, endpoint, model and key.
- When you use an external model, the data in each AI request is sent to the provider you configured.
- Which parts of a sheet a request includes depends on the tool. Exactly what each request contains will be documented before release.
- A proposal comes back and is shown as a preview. Nothing is applied until you approve it.
Keys and settings
- Your key is held only in memory while the app runs. There is no secure keychain integration yet.
- The settings file never stores keys, sheet contents, AI proposals or approval records.
MCP
MythSheet has MCP contracts for reading, analysing and proposing. Writing changes through MCP is blocked until a trusted approval channel is complete.