coreCerebrum gives everyone one shared brain, gives every task its own running agent, and shows what each piece of work cost. People stop re-solving yesterday's problems and start shipping today's.
30-day free trial. Card on file, first charge in month 2. Cancel anytime before then.
Most teams run AI the way they started: one chat window per person, one agent per repo, and a bill that only shows up at the end of the month. That is where the money leaks out.
Stop the leaksSomeone works out a hard problem. Next week a teammate asks the same thing and pays for it from scratch.
Conventions, gotchas and decisions vanish when the window closes. Every session rediscovers them.
Every turn re-sends the whole context, so a long session costs more with every reply and nobody sees it happen.
The agent is the bottleneck. You sit idle while it thinks instead of starting the next task.
Six pieces that work together. Pick one to see the real screen behind it.
Decisions, conventions and gotchas are captured while people work, cleaned up on their own, and recalled on the prompt that needs them. One person solves it, everyone keeps it.
router: 3 memories relevant to src/auth/
gotcha guard runs before the ORM check
decision tokens verified server-side, never client
bug bug-114 session fixation, fixed 2026-06
surfaced before the first file was read
Every card on the board is a real agent session in its own git worktree, on its own branch. Start five at once, watch them move lanes, step in when one needs you.
A no-terminal surface for the rest of the team: draft email, marketing, security, dev requests, proposals and Design DNA, all drawing on the same team memory.
Your Gmail inbox inside the board. Archive, label and reply, or turn a thread into a card with the task already worked out for you to confirm.
Inbox
Convert to card
Inferred task
Fix the failing checkout webhook retries before Friday
taskreplycustom
Shown before the card exists, so you can correct it.
Create cardGitHub and GitLab issues, Teamwork tickets and time, Clockify, Google Search Console, Google Ads, WordPress deploys, Cursor and MCP servers.
A warning before a session gets expensive, re-read detection, and spend by project and person. You see the cost while it is still cheap to change.
This session is getting expensive
Refactor auth guard is carrying 412K tokens of context.
Every turn re-sends all of it, so each reply costs about 8 times what it would in a fresh session. Compacting summarises the history and starts the count again.
Write a card, get a branch. coreCerebrum cuts a fresh git worktree for every card and starts a real Claude Code session inside it, with the harness loaded and the team memory already recalled.
Backlog2
You write the card
In progress2
An agent works in its own worktree
Review1
You check a normal branch diff
Done1
Merged and cleaned up
Each worktree is a real directory on a real branch that shares the repository's object database, so it is cheap to create and nothing lands on main by accident. When an agent needs a decision, its card moves to Needs you and stops.
How orchestration works$ git worktree add ~/.corecerebrum/worktrees/11fd0 -b cerebrum/11fd0
worktree created, HEAD at cerebrum/11fd0
harness loaded: 24 skills, 22 commands, 10 hooks
memory recall: 3 prior decisions for this path
session live
Team memory is the part most agent tools leave out. Say "we decided", "the convention is" or "gotcha" while you work and it is saved. A newer decision supersedes the old one, history is never hard-deleted, and the right memories are recalled at the start of every session.
router: 3 memories relevant to src/auth/
gotcha guard runs before the ORM check
decision tokens verified server-side, never client
bug bug-114 session fixation, fixed 2026-06
surfaced before the first file was read
Running agents all day turns token spend into a real line item. coreCerebrum watches it for you and speaks up before the invoice does.
This session is getting expensive
Refactor auth guard is carrying 412K tokens of context.
Every turn re-sends all of it, so each reply costs about 8 times what it would in a fresh session. Compacting summarises the history and starts the count again.
Admin, usage overview
7 days30 days90 days
Not all of the work is code. Cowork is a no-terminal surface for sales, marketing, ops and owners, running on the same board and the same memory as your developers.
Researches the recipient, drafts the reply and saves it to your Gmail drafts.
Blog posts, SEO meta, ad copy and social, with live word and keyword counts.
A topic, competitor or question turned into a brief with clickable sources.
Business asks land in a Requests queue for a PM to triage onto a board.
Inbox
Convert to card
Inferred task
Fix the failing checkout webhook retries before Friday
taskreplycustom
Shown before the card exists, so you can correct it.
Create cardGmail lives inside coreCerebrum. Archive and label it, reply from it, or convert a thread into a card. The thread is read once and turned into a concrete instruction you can see and correct before the card exists.
Email managementWe have no customer logos or testimonials to show you yet, so there are none here. Here is what you can check for yourself.
The screens on this page were captured from a running install. Only project and card names are changed.
Usage is read from the per-turn transcript your AI tool writes, then attributed to the session that spent it.
Destructive operations are refused by a hook before they run. The model is never trusted to remember.
The 'corePHP' team runs its agency on coreCerebrum, and corecerebrum.guru itself was built by a card on the board.
The same AI you already pay for, with a team layer around it.
| The job | Ad hoc chat windows | With coreCerebrum |
|---|---|---|
| Where solved work goes | Stuck in one person's chat history | Team memory, recalled on the prompt that needs it |
| Running work in parallel | One window at a time while you wait | Many cards at once, each on its own branch |
| Keeping agents safe | A prompt asking the model to be careful | Hooks that refuse destructive operations before they run |
| Knowing what it cost | The invoice at the end of the month | Warnings mid-session and spend by project and person |
| Reviewing the work | Whatever got pasted back into the chat | A review gate and a normal branch diff |
| Email that needs action | A separate tab and a copy-paste | Threads become cards with the task inferred |
| The rest of the team | Separate tools and separate logins | Cowork on the same brain: email, marketing, research, Design DNA |
| Library documentation | Whatever the model remembers from training | A docs service that crawls the current docs |
Every plan works with the AI you already use. You keep paying your AI provider directly; coreCerebrum makes that spend go further.
$30per user/month
Billed monthly. Billed only for active seats.
$25 per user/month at 100+ seats
For anyone who uses AI to write, research or answer questions.
Most popular with teams
$100per user/month
Billed monthly. Billed only for active seats.
For developers and technical builders.
$200per user/month
Billed monthly. Billed only for active seats.
For people who build software and websites.
A custom coreCerebrum built inside your own infrastructure: on-prem, private cloud, AWS GovCloud or Azure Gov. Custom seat pricing, SSO, security review and a named contact.
Quoted
Talk to usAll self-serve plans include the 30-day free trial. Card on file, first charge in month 2. You are billed only for active seats (a seat that did not use coreCerebrum that month costs nothing). Save 20% with annual billing. Full pricing and seat calculator
Anything else? Email [email protected] and a person from the team replies.
All questionsYes. You keep your own plan or keys and keep paying your AI provider directly. coreCerebrum never resells or marks up your AI usage. Its job is to make that spend go further.
Your chat app remembers one person's preferences. Team memory keeps your whole team's solved work: decisions, conventions and gotchas are captured as you work, de-duplicated, and recalled on the prompt that needs them, so when one person figures something out, everyone can reuse the answer.
Memory ($30 a user) is team memory and lighter requests for anyone who uses AI to write, research or answer questions. Build ($100) adds agents that run side by side, each in its own safe workspace, plus the developer toolkit. Studio ($200) adds Design DNA for websites, landing pages and presentations, with export and one-click publish.
The board runs locally against your own checkouts, and every agent works in a git worktree that is a normal directory on your disk. The local server only listens on your own machine. When the optional team backend records usage, it sends totals such as token counts and cost, never transcript content.
Several ways at once. Solved work is recalled from team memory so it is not paid for again. A session that has already read a file is told so before reading it again. A session whose context has grown large raises a warning with the option to compact it. And spend is shown by project and person, so you can see where it goes.
Nothing for 30 days. Your first charge is on day 31, for your active seats. Cancel before then and you will not be charged.
Start with a Memory seat at $30 a person, add Build for the people who run agents, and see the difference on your own work. No charge for 30 days.
No charge for 30 days. Your card is billed starting month 2.