Teach it once. From then on, it speaks for you.
How Brain Console turns your knowledge into answers your whole team can use.
Write your brain in a plain editor.
You write your brain in a plain editor – about the business, your voice, your services, your clients, how things actually get done. It stays exactly what you write it to be.
"If you can write a doc, you can do this. No setup, no schema, no consultant."
An employee asks. The answer comes back in your voice.
Every question from the team gets answered from the brain files you wrote. The system finds the relevant pages and uses exactly that to respond – not what it learned in training, but what you said.
> How do we handle a scope change request from a client mid-project?
Put it in writing first. Acknowledge the request, then follow up within two business days with a revised timeline and cost estimate. Don't agree on scope verbally – use the change request form and get sign-off before any work starts.
`process/scope-change.md``voice.md`The answer comes from what you wrote, not from what the AI guessed.
Your brain gets broken into pieces and stored by meaning, not keywords. When a question comes in, the system finds the pieces that are actually about that question and hands only those to the AI to answer from. That's why it answers from your documents instead of making things up, and why every answer points back to a source. The model isn't remembering your business – it's reading the right page of it, every time.
Every response is built from what you wrote in your brain, not from model memory.
Each answer links back to the exact brain file it came from.
Per-tenant data separation enforced at the database layer with row-level security.
Any action that changes something – a reply, a log, a task – waits for your explicit approval.
Beyond answering, it acts.
Brain Console can draft the reply, log the follow-up, pull the record. Every action that changes anything waits for explicit approval.
Nothing happens without your yes.
“Thank you for reaching out. Our typical onboarding runs 3–5 business days from contract signing. I'll send the full timeline document this afternoon.”
`clients/acme.md``services/onboarding.md`Connectors are rolling out with our design partners – tell us which tools matter.
A skill is an SOP the AI actually follows.
A skill is not a document a human reads and then improvises from. Write the skill once. The team runs it on demand. The AI runs the steps.
Every time. Not approximately. Not with a different interpretation each run.
- Check for an existing record on the contact's domain.
- Draft a reply using the voice guidelines in voice.md.
- Log a follow-up task in the pipeline with a 3-day window.
- Propose a meeting slot from the team's shared availability.
For technical evaluators.
The specifics, without the glossary.
- retrieval
- Hybrid dense vector search plus BM25 keyword search, fused with reciprocal rank fusion (RRF). The best of both methods, not a choice between them.
- indexing
- pgvector with HNSW indexing. Fast approximate nearest-neighbour search that scales without degrading.
- citations
- Per-source grounding on every answer. Every response traces back to the brain file it came from.
- prompt caching
- Repeat context served from cache. Fast, low-cost responses on recurring queries without re-loading the same context each time.
- data isolation
- Per-tenant isolation enforced at the database layer with row-level security. Not in application code. The boundary holds even if application logic has a bug.
- foundation model
- Built on Anthropic Claude.