Let ChatGPT, Claude or Claude Code read what Yaven knows
Yaven is a memory layer any AI can use. An MCP server hands your assistant of choice the context you allow — per-agent keys, scopes, receipts, Stop.
What an agent can read
- yaven_about_me — your understanding by area.
- yaven_people, yaven_who_is — your people and what Yaven knows about one of them, with how sure it is.
- yaven_upcoming, yaven_timeline, yaven_promises, yaven_library — the calendar, the storylines, the promises, what you saved.
- yaven_remember — only with a write grant; stored as an agent's candidate, unconfirmed, fading after 60 days unless you confirm it.
Answers are written by code from your data — no model call on your credits. Health, money and beliefs are left out unless you turn them on for that agent.
Control
- One key per agent, shown once, stored as a hash.
- Scopes per agent; a write grant is separate.
- A receipt for every call, refusals included, kept 30 days; 120 calls per agent per hour.
- Stop revokes at once. Me → Connections → Yaven for your AI.
Honestly:Streamable HTTP, key in the header or in the address. No OAuth yet, so connectors that require it cannot use it; Claude Code and any MCP client that takes a key can.
Questions
Can the agent change my data?
Only if you grant write, and even then it can only suggest things to remember; they are marked as coming from that agent and are never treated as facts until you confirm them.
Free to start. Say the first thing out loud.
Yaven is on Google Play. Every screen is free; AI work runs on credits you can earn, redeem or subscribe for.
