search
The memories that best match a query, ranked, with no judgement on whether they bear on it.
for m in mem.search("pricing", limit=5):
print(m.text)for (const m of await mem.search("pricing", { limit: 5 })) console.log(m.text);Search makes no model call, so it comes back in a fraction of a second. It ranks by position, not by confidence:
it returns its best matches for any question at all, including one the memory knows nothing about, and
there is no score in the results that means "this is relevant". If you need to know whether anything
bears on a question, use context or ask.
Lessons, stories and promises
Geniffy learns more than facts from what is saved: rules and how-tos (lessons), what happened (stories, called
episodes), and promises with their due dates (intentions). Search finds those too, by meaning and by the words
themselves, so "port 6543" finds the how-to that holds it. search_full returns them beside the memories:
found = mem.search_full("connection limit")
for lesson in found["lessons"]:
print(lesson["statement"], "when:", lesson["applies_when"])
for promise in found["intentions"]:
print(promise["what"], "due", promise["at"])const found = await mem.searchFull("connection limit");
for (const lesson of found.lessons) console.log(lesson.statement, "when:", lesson.applies_when);
for (const promise of found.intentions) console.log(promise.what, "due", promise.at);Over the API, POST /v1/search returns lessons, episodes and intentions beside memories, each naming the
source it came from. With a label filter, or a kind other than all, only memories come back.