harnyUse case
Memory for an internal knowledge base
An internal assistant answers from documents, chats and meeting notes written by many people over years. Memory for a knowledge base has to keep the newest statement in front, keep the older ones answerable, and name the document behind each answer.
What you are building
An assistant that answers employee questions from the team's own writing, and shows where each answer came from.
Why memory decides whether it works
- Documents contradict each other. The wiki says one thing and last week's chat says another.
- Old pages look as sure as new ones. Text that looks like the question is not the same as what is true now.
- People need the source. An answer without a link is one more thing to check by hand.
What Harny provides
- Raw text with its original timestamp, from any source you can read.
- Recall that returns current facts with dated evidence and source labels.
- No schema to define, no embedding model to choose and no chunking pipeline to build.
Getting there
- Send each document or message to
POST https://api.harny.ai/v1/memory/ingestwith an id you choose, a label and its timestamp. - Poll
GET https://api.harny.ai/v1/memory/ingest/{ingestionId}until the status is completed. - Ask
POST https://api.harny.ai/v1/memory/recalland show each result's source label beside the answer.
Frequently asked questions
Do I have to split documents into chunks first?
No. Send the text. Harny prepares it for recall.
What if a page is deleted at the source?
Call DELETE https://api.harny.ai/v1/memory/data-points/{sourceId} with the id you used when you sent it.
Keep reading
Go on. Get Harny.
The first 500 monthly-active end users are free for good. Create an app key and make your first call in minutes.