Mission
MCP
Talk with any AI or any person. Keep what matters in the brain. Every other AI and every teammate can recall it. Knowledge moves with you.
Cursor@thinkingroot what did we agree?
Claude · MCP
api.thinkingroot.com/mcp
$40 a seat. Pilot October 1.
Console
Browse everything added. Surface what is critical. Ask root when you need the answer.
Chrome
Type in any panel. If that person is in the brain, shadow text brings back what you already did with them. One click records any meeting, solo or in Google Meet, in high quality, then puts that context back at your cursor.
Shadow text
Type a name. If they are in the brain, the past lights up at the cursor.
Record
One click. Solo or in Google Meet. High quality, context kept.
At the cursor
That context reaches you again wherever you write next.
Tools and chats both land in one brain. Filed where they belong.
Tools
Docs and apps sync in
Notion, Drive, and the rest land in the right branch for each client or project.
Chats
Conversations stay too
In Claude or Cursor, a trigger word saves it. In Slack or the compose box, one tap keeps it.
The same memory shows up wherever you work. AI tools and the browser.
AI tools
In every AI tool
Paste one URL. Claude, Cursor, and Codex already recall what the company knows.
Browser
TabIn every tab
A pill at the edge of Gmail, Slack, or any page. Tab to put the line into what you are typing.
Time back. Longer sessions. Cleaner answers. Less spend on tokens that never earned their place.
Time
The answer is already there. No hunt through threads, docs, or last week's chat.
Long sessions
The thread can run for hours. The brain still knows who you meant and what you promised.
Output quality
Replies draw from what the company already decided, not a fresh guess every turn.
Productivity
Less paste, less re-brief, less starting over. Work moves from the question.
Near zero hallucination
If the brain does not know it, it stays quiet. No inventing to fill the gap.
Token reduction
Only the facts that earn a place ride into the model. Shorter context, lower cost.
Type a line. The fact that matters rises. The rest stays quiet.
What rises at your cursor
seeded demo brain
You quoted $40 a seat on Thursday's call.
ConflictYou typed $45. Brain has $40.
Postgres over Mongo: one writer, no lock class.
I'll send the terms Friday.
Acme starts October 1.
Priya is out until the 18th.
Dana, VP Ops, is Acme's champion.
What you typed disagrees with what was said. The brain rose the record.
Same question to a file, a repo, and the brain. See who answers, and what you paid to read.
notes.md · 28 Aug
≈ 569 tokens
Whole file rides in, every ask.
grep -ri 'quote|acme'
7 word hits
Still strings. You read them to know.
recall · ≈ 38 tokens
≈ 38 tokens
Only what answers. Nothing else.
World #1 on LongMemEval-S. Zero memory-poison hits on RotBench. NVIDIA Inception member. Built like a human brain, scored on the production path.
Every claim on this site is measured or labeled as pending. Same rule here.
Context for humans and AI tools, at the cursor where they already work. Sync what lives in your systems. Keep what lives in heads and chats. Hand it back on every seat: Cursor, Claude, Chrome, and every teammate you add.
Docs, mail, calls, and AI chats do not share. Each tool starts over. What lives in one head dies when the tab closes. Agents and people need the same past. A chat cannot hold that. A folder cannot serve it at the cursor.
A file is storage. It does not show up in the composer you are already typing in. GitHub holds code and docs. It does not catch mail, meetings, or what someone just decided in chat. Searching a repo is not the same as asking what you agreed, on the branch you chose.
Paste MCP into Cursor or Claude and ask in plain language. In Chrome, shadow text can finish the line in the box you are already in. On Meet or in person, the pillar Record button files the call on the branch you selected. Same brain. Every tool.
World #1 on LongMemEval-S at 99.0% Recall@15 on the official 500-question suite, measured on the production path. Zero memory-poison hits on RotBench. Method write-ups are on the blog. We do not invent ranks we have not run.
No. Isolation is architectural, not a filter. Each user's memory is a separate database. Requests are scoped at the gateway. Embed tokens pin a single user. There is no shared table for data to cross.
No. Reach it over MCP, SDK, or REST with the models and tools you already use. Free tier to start. Pro, Scale, and Business as you grow. Your memory is not trapped inside one vendor's chat product.
ThinkingRoot achieves 99.0% Recall@15 on LongMemEval-S - the highest published retrieval accuracy on the benchmark - with only 481 tokens context and 16ms warm latency. Full methodology, graph architecture, and head-to-head comparison against Supermemory and Zep.
Read the post →
Every memory product degrades as it accumulates - and nobody measures it. We built the first degradation-over-time benchmark, ran it on our own engine, and published the times it caught us being wrong.
Read the post →Sync what lives in your tools. Keep what lives in chats. One company memory, back at the cursor in every AI tool and every seat.
