Search what an agent has saved

Search looks through files and notes once they're indexed. It doesn't search plain key-value data. For developers searching what an agent saved. Pro is $9 a month.

Document search

POST /api/rag/search takes a query, an optional namespace, and an optional topK. topK defaults to 5 and is capped at 20. The call meters one rag_query. Omit namespace to search every namespace the key may access.

MCP search_documents with source=documents is the same search. The response includes hits plus budget and, when the key has one, keyBudget.

Wait until the file status is ready. A search issued while the queue is still embedding will not contain that upload yet.

curl -s -X POST "https://tillpad.dev/api/rag/search" \
  -H "Authorization: Bearer tp_…" \
  -H "Content-Type: application/json" \
  -d '{"query":"email before noon","namespace":"jobs","topK":5}'

Memory search

POST /api/memory/search takes a query, an optional scope of prefs, facts, or run, and an optional topK. It meters one rag_query. MCP search_documents with source=memory checks the memory_search allow-list name and can filter by scope.

Exact key lookup is GET /api/memory/:scope/:key or manage_memory action=get. That path meters kvp_ops, not rag_query.

Questions people ask

How many search hits come back?

topK defaults to 5. Values above 20 are capped at 20. Values below 1 are raised to 1.

Does search read every key in a namespace?

No. Semantic search uses the vector index built from files and notes. Key-value listing is manage_kv action=list or GET /api/kvp/:namespace.