The problem with a brokerage knowledge base
The usual version is a folder of documents and a search box. It goes stale the moment a policy changes, because the document describing the rule and the system enforcing the rule are two different things, and only one of them gets updated.
Bolting a chatbot onto that folder makes it worse, not better: now a confident answer can be wrong, and nobody can tell which sentence came from the handbook and which the model made up.
The rules that run are the first source
Raven already knows your document requirements, which ones block a close, which stage each is due at, what your task lists create, and which compliance rules are active. A question about what is required on a listing is answered from the checklist that will actually be enforced on the next transaction, so the answer cannot drift from the behaviour.
- Answers drawn from your live checklists, task lists and compliance rules
- Your reviewed state defaults, with the citation each item was approved against
- Personal checklists and task lists an individual created are excluded; this is the brokerage's policy, not one agent's
The first layer is what the platform will actually enforce on the next transaction, so the answer cannot drift from the behaviour.
Your own material, for everything a rule cannot express
Upload the handbook, the recorded training, the onboarding video, or point at a page. Raven reads documents, transcribes audio and video, and crawls a single page, then answers from the passage that actually addresses the question, with the source named.
- PDFs and documents, read with the same extractor the platform uses on contracts
- Audio and video, transcribed
- A single web page, fetched with robots respected
- Passages that do not answer the question produce “I don’t know”, never a confident guess
Practice is labelled practice. An answer from your brokerage’s own habit is never presented as a legal requirement.
Who can see what, and what is actually verified
Knowledge is scoped the way your brokerage is: company-wide, one office, one team, or the person who uploaded it. A source outside a reader's scope is absent from their answers rather than redacted, and that applies to brokers too.
- Visibility by organization, office, team or owner
- An answer from your brokerage's own practice is labelled as practice, not as a statutory requirement
- Re-uploading the same file keeps the audience it was narrowed to; silence never widens access
- Every upload, change and removal is recorded
A source outside a reader’s scope is absent from their answers, not redacted in them. Brokers included.
Frequently asked questions
How is this different from uploading a PDF to a chatbot?
Two ways. The first source of truth is the rules the platform already enforces, not a document describing them, so the answer cannot drift from what actually happens on a transaction. And an answer that comes from your brokerage's own practice is labelled as practice rather than presented as a legal requirement, which is the distinction that matters when an agent is deciding what they have to do.
Can an agent see another office's policies?
Only if the source was published to them. Knowledge is scoped to the whole company, one office, one team, or the uploader, and anything outside a reader's scope is simply absent from their answers. The same rule applies to brokers.
What happens when EVE does not know?
She says so. A passage that does not answer the question is not turned into a confident answer, which is the failure mode that makes a knowledge base untrustworthy after the first wrong answer.
What can I upload?
Documents and PDFs, audio and video recordings, and a web page by URL. Recordings are transcribed. There are per-file and per-day limits so a single upload cannot run up a transcription bill.