Everything your agents read, searchable later
CrawlGenie Memory is AI agent memory you can inspect. Most teams read the same sources every day. CrawlGenie Memory keeps what was read, in collections you control, and answers the next question from it: passages with sources, a few hundred tokens, no re-reading. It is plain full-text search on your own workspace, which means you can see exactly why an answer came back.
~300 tokens per question
Instead of re-reading a 48-document corpus, the model gets the five passages that match.
Inspectable
Every collection, document and passage is visible and deletable. No vectors to explain.
Fed by everything else
A crawl, a monitor run, a document or a page can be sent to Memory with one click or one call.
What happens when you use Memory?
Create a collection
"Competitor docs", "Our policies", "Q3 filings".
Add material
URLs, files, a finished crawl, or the rows a monitor collected. OCR runs when needed.
Ask
From the dashboard, ChatGPT, Claude or your code: passages, sources and a receipt.
What do people use Memory for?
Support
The help centre, crawled once, answers tickets from passages.
Sales
Competitor documentation at your fingertips with sources for every claim.
Research
Hundreds of filings and papers, queried for a few hundred tokens each.
What you get, in detail
Agents forget between runs, and people forget between weeks, so the same documentation, filings and competitor pages are fetched and re-read constantly. CrawlGenie Memory keeps what was read in collections you control: a crawl of a help centre, a folder of PDFs, the rows a monitor collected, or pages an agent saved mid-task. Material is split into passages with their headings and indexed with full-text search in your workspace, so a later question returns the five or six passages that match, each with its source, and a receipt showing how many tokens that replaced.
There are no embeddings and no vector database to run or explain: retrieval is exact-word search with BM25 ranking and stemming, which is strong for "where does it say" questions and honest about its limits with paraphrase. Collections are visible and deletable document by document. Memory is fed by the other products with one click or one call and is available to ChatGPT and Claude through the connector and to agents as two tools, so the cheapest answer is always the one that was already read.
For developers
POST /v1/knowledge, /documents, /search; MCP tools knowledge_add and knowledge_search. How retrieval works.
What would this save you?
Move the sliders. The saving is computed from the measured ratios above and the model price you pick.
Assumes 10% of the page is kept per read (measured range 5–15%). Receipts in the product show the real number for your traffic.
Questions people ask
Does it use embeddings or a vector database?
No. Full-text search with BM25 ranking in your workspace; strong for "where does it say", weaker for paraphrase, which the docs say plainly.
Who can see my collections?
Only your workspace or organisation. Built-in product docs are the only shared collection.
Spend tokens on answers, not on pages.
Create a workspace in a minute. Connect ChatGPT, Claude, Cursor or your own agents. Every answer comes with a token receipt.