About
The AI is smart. It just does not know your company.
Every company runs on knowledge that lives nowhere in particular. How you deploy. Who owns which customer. What was decided in that call on Tuesday, and why. It is spread across threads, docs, tickets, and people’s heads, and no one person holds all of it.
When a company hires someone, they spend their first month just finding things. That is an expensive but familiar cost. What is new is that companies are now also running AI agents that face the same problem, except worse: they start every session knowing nothing at all. So the same context gets pasted into prompts over and over, and the answers stay generic — because generic is all they can be.
Nobody was building the layer in between. Companies had knowledge tools written for humans, and they had AI agents, and nothing translated one into a form the other could actually use. That is what Indexbrain is.
Decisions
What we have refused to build
Scope is the hardest part of a product like this. These are the things we were asked for, thought about properly, and deliberately said no to.
We do not connect to your database
It would be the fastest way to make demos impressive and the fastest way to become a serious breach surface. Indexbrain reads through the APIs of tools you already trust, and nothing else.
We do not keep your raw content
Source material is purged on a rolling 90-day cycle and AI conversation transcripts go as soon as knowledge has been extracted. Holding onto everything forever would make some features easier. It is not worth it.
There is no auto-approve
Not as a setting, not on an enterprise plan, not as an override. Every write to an outside system stops for a person. We were asked for this early and said no, and the answer has not changed.
We are not trying to be your agent
Claude, Cursor, ChatGPT, and Codex are extremely good and people already like them. Building a sixth chatbot to compete would have been a worse product and a worse business. We are the layer underneath.
How we work
Four things we hold to
Not values on a wall — these are the rules that decide what gets built and what gets deleted.
Verify, then claim
Everything on this site maps to something actually shipped. Where a capability is behind a flag or not yet enabled in production, we say so — including on the security page, which lists what is not built alongside what is.
Delete more than we add
Apps, Workflows, and a first version of skill execution were all built and then removed. Two structurally identical systems became one. A feature that is not carrying its weight is a liability, not an asset.
One phase at a time
Small pieces, landed and confirmed, rather than a large feature dropped in one pass. It is slower to describe and considerably faster to trust.
Cost discipline is a feature
Processing is batched and gated so a credit buys real throughput rather than being burned on noise. That is why the free tier can be genuinely free instead of a two-week trial wearing a disguise.
You can see the practical version of all of this on the changelog — including the parts that were built and then taken back out.
Come and kick the tyres
The free tier is not a trial. Connect a source, ask it something only your company would know, and judge it on the answer.