A document
Add a page or a PDF by its address, or paste the text of a document. It is kept whole in the library, where questions can find and cite it. Give code a title that ends in the file's name (store.c, tool.py) and it is indexed as code.
Pick a profile to see what the digests say about it and what it links to.
Server Settings
Your account
Signed in as ().
Sign in a phone
Shows a QR code for the iPhone app: point the iPhone's Camera at it and the app signs in as you, with no name or password to type. The code works once, for 5 minutes.
Profiles
How many digests must name a company, person or agency before its profile is listed for you. Searching Profiles by name always finds every one, and researched or hand-added profiles are always listed.
Change your password
API tokens
For things that cannot fill in a sign-in form: the iPhone app, an MCP client, a calendar subscription (add &token=… to the feed's address). A token acts as you, with your topics, until revoked. It is shown once, when made.
Audit
Every sign-in attempt, and everything signed-in people do in Cerebrum, newest first. Passwords, tokens and keys are never written down. The page's own background checks are left out, and the same page opened again by the same person within ten minutes counts once. Admins can sign in only from inside the Tailscale network.
Sign-in attempts
Activity
Users
An admin sees and does everything. A user sees only the topics ticked for them -- their digests, and only what those digests say in Profiles, the Calendar, Reports, Search and the chats -- and no Settings beyond their own account.
Research
What a profile's Research asks for. Research sends the profile to its topic's research webhook (Settings → Plugins → Topics); these settings go with it, for what answers the webhook to use: the sites to search first, words added to the search, and what to look for first. Pick a topic on the left. A new plugin shows up here by itself.
All topics
Sites it never reads
Sent with each Research request as sites never to read, with their subdomains, one a line.
Acronyms
Each acronym spelled out, linked to the profile with that full name. A digest or a research that says ONR then joins Office of Naval Research, and a profile called only "ONR" is folded into it. With no profile of that full name yet, the one called only "ONR" is renamed to it. "Not this one" takes an acronym off a profile it was wrongly given to. Changes apply at once.
Type the full name; profiles with names like it are suggested as you type.
Set here
Called only by an acronym
Agencies and companies the digests or research name only by an acronym, the most named first. Link one to its full-name profile.
Found in digests and research
Emblems
Every seal, logo and flag Cerebrum has, once each, the most used first. Rename one and every profile showing it shows the new name. Replace one and they all show the new picture. Remove one and they show no emblem (and are not looked up again; set one from the profile's Change emblem). An uploaded picture can be picked from any profile's Change emblem.
About
Checking...
Version —
Last commit —
Embeddings: search and relate by meaning
An embedding model turns each passage of the library and each idea into numbers, so that texts with the same meaning lie close together. With one, the language model can also look for passages that say what a question asks in other words, ideas are related by what they mean, and ideas that are one idea under two names can be found. It is another server in the same format as the language model (Ollama, for example, with nomic-embed-text). The search by words stays, for exact names and numbers.
Prefixes
Embedding models are trained to see a short tag before the text: nomic-embed-text wants "search_document: " before what is kept and "search_query: " before a question. The usual tag of the model's family is used unless you write one here; write none for no tag. Changing the tag for passages makes the vectors again.
What's running
Checking...
Keys
Keys the brain uses to learn. They are written to the server and never shown again.
Firecrawl fetches the pages you add by address, and reads them better than a plain download does. Brave finds candidate pages.
Grok
Connect Grok by OAuth. You approve it on xAI's page with the code shown; the brain keeps and refreshes its own tokens.
What's connected
Checking...
Grok: checking...
New topic
Every topic the left column shows gets its own documents, Profiles, Calendar and Activity.
Topics
Checking...
The address to paste into whatever speaks MCP for each topic. Each topic's digest sender has its own token, made with the topic and shown once: it can only save documents to that topic, from anywhere. Give the sender the address with its token, or the token as an Authorization: Bearer … header.
Heartbeat and limits
How often the heartbeat ticks: on each tick it reads the next address on the reading list, if there is one. Lower means more frequent activity; higher means slower, calmer work.
The largest document Add a document's file picker accepts, in megabytes.
Profiles
How many of a topic's digests must name a company, person, agency, solicitation, contract or event before it gets a profile in that topic's Profiles. Below this it is only hidden, not deleted: it still appears from its digests, its dates stay on the Calendar, and it shows up by itself once enough digests name it.
What's set
Checking...
Checking...