AI connectors
Search public hypotheses without signing in. Connect your account to review and save useful assumptions, evidence, and decisions, then resume that project in a later conversation.
Decision memory
Search by problem or audience
Fetch the full decision record
Compare evidence in a focused widget
One MCP endpoint works with both clients. The public library needs no account; private project memory uses OAuth and scoped permissions.
In ChatGPT Settings, open Security and login and enable Developer mode.
Open Plugins, choose the plus button, and add the MCP endpoint shown above.
Enable Hypotheses in a new chat. Public search works immediately; a private-memory request opens secure account authorization.
In Claude, open Customize → Connectors and choose Add custom connector.
Name it Hypotheses and paste the public MCP endpoint shown above.
Enable it from Search and tools. Public search works immediately; private memory shows an inline Connect step.
Search, fetch, and comparison still read only published briefs. Private tools can list and resume only the authorized account's projects. Saving requires a separate preview followed by an approved, signed write request; the connector never imports a full conversation automatically.
Private memory beta
Ask your AI to prepare a Hypotheses capture. You review the exact project, exploration, branches, assumptions, evidence, and decisions before the write tool can run.
Sign in to Hypotheses and grant separate read and save permissions.
Inspect the exact structured capture and change anything that is wrong or too broad.
Approve the client write prompt. Signed previews expire after ten minutes and retries are idempotent.
Ask in natural language and continue reasoning from stable public record identifiers.
Every result links to its canonical public Hypotheses page for inspection and citation.
Public tools read only published public records. Private memory requires account authorization, and every save requires a reviewed preview and separate approval.
Starter prompts