No tagging
Nobody tags or files anything. Arrival is enough.
How it works
Content arrives. Questions land once.
Nobody tags or files anything. The reading is done before anyone asks.
01
It is read when it arrives.
When content arrives, Cluesora reads it, summarises it, connects it to everything related, and checks it.
Two summaries
Every page gets a short summary and a longer study summary. Both rewrite when the content changes.
Bad input held
Garbled scans are held back until fixed. A passage repeated fifty times counts once.
It checks itself
After every big change it checks its own work and fails loudly instead of quietly getting worse.
Your workspace · permissions follow every source
Ways in
- Upload — PDF, Markdown
- Chrome clipper — only what you Save
- Confluence import — copy with a link back
- AI agents (MCP) — Claude, Cursor, Codex, Devin, any MCP agent
- Teammate — app, Slack, Teams
- Jira · email · code hosts — planned
arrives
Read on arrival
Same checks for every way in
- Read and summarised (short + study summary)
- Connected to everything related
- Checked — garbled scans held back, repeats count once
- Permissions attached from the source
kept
The map (Corpus)
- Pages, with their sources
- Ideas and what they depend on
- People, systems, customers, vendors
- Tables kept as rows
- Every version kept, dated
used
Put to work
- Search by intent → exact line + why it matched
- Teammate answers, with sources
- Agents search first — reads never metered
- Fathom — courses, checks, vivas
Dead ends and agent updates are written back
02
Many sources, one answer.
When several documents say the same thing, you get one answer with all of them cited.
Extra detail kept
What one source adds, like an edge case or a caveat, is kept.
Two books, one idea
A new handbook on load balancing joins last month’s book. The idea now has two sources. Progress stays.
Conflicts stay split
One page says refunds take 14 days, three others say 30. You see both sides and where each came from.
Before
Two handbooks on the same idea sit in two folders. Progress on one does not count for the other.
After
The idea has both sources. Anyone who already studied it keeps their progress.
03
Answered by intent.
The question is matched to what your organization actually decided and meant, not to the words on a page.
Words still work
It also matches keywords and meaning, but intent is the difference.
The billing question
“Can a new enterprise customer go monthly?” finds “Why we bill annually.” The question never says annually.
What it meant
Enterprise pays yearly. Customers on monthly plans left at renewal. Billing cannot prorate mid-term.
Asked“Can a new enterprise customer go monthly?”
Why we bill annually.
Enterprise customers pay yearly.
- Why
- Customers on monthly plans left at renewal.
- Because
- Billing cannot adjust partway through a term.
- Matched
- By intent. The question never says “annually”.
04
The answer shows its work.
Every result shows the exact line, its source, why it matched, and what it connects to.
Exact line
You see the sentence that answers you, not a page to go hunt through.
Why it matched
Each result says why it matched: by intent, by meaning, or by the words.
What it connects to
Related ideas, people who can answer, and the sources sit next to the line.
- Line
- “Enterprise customers pay yearly.”
- Source
- Confluence · Pricing policy
- Why it matched
- Intent: enterprise billing terms
- Connects to
- Pricing review · Proration limits · Billing team
05
The story is already mapped.
When the story is spread across several places, Cluesora follows connections it has already mapped. It does not guess the route.
Still one hop
The reading is done before anyone asks, so the path is ready when the question arrives.
No page wandering
You do not watch an agent walk from page to page hoping the next link helps.
Caveats travel
An edge case living in a second document arrives with the main answer, cited.
- Pricing policy
- Exception memo
- Billing FAQ
06
People, systems, and order.
It knows the real-world things your content mentions, and what must be understood first.
Things with names
Systems, services, teams, customers, vendors. “Which team owns this service?” is a question that works.
People who can answer
It knows who can actually answer, not just who is mentioned.
Learn in order
For every idea it knows what must be understood first, so learning follows the order ideas depend on.
Progress stays put
Ideas keep their identity as content changes, so progress is never lost when new material arrives.
- Idea
- Enterprise billing
- Learn first
- Plan types · Proration rules
- Owned by
- Billing team
- Can answer
- Finance lead
07
What the library assumes.
When something your content needs is not taught anywhere yet, Cluesora keeps a list.
Assumed, never taught
It shows exactly what your library assumes but never explains.
Listed once
Fifty topics assume basic Python and nothing teaches it. Cluesora lists it once.
Links fill in
When a Python primer is added, all fifty connect to it.
- Needed
- Basic Python
- Assumed by
- 50 topics
- Taught by
- Nothing yet
- After a primer is added
- All 50 connect to it
08
The map stays correct.
Routine upkeep is automatic. Unclear calls get checked before they change anything. Anyone can correct what is wrong, and their edit wins.
Dead ends write back
“What’s our SLA for enterprise P1s?” hits a wall. Teammate asks the support lead in Slack. Their answer is kept.
Agents extend, not fork
Agents update the existing doc instead of filing a second version. Sources are never rewritten.
Versions kept
Everything is versioned. New versions take over. Old ones are kept and dated, never deleted.
About a minute
Small updates are usable within about a minute. Large PDFs take up to ten minutes.
Before
An agent writes a second procedure beside the real one. Next week nobody knows which text is true.
After
The existing procedure is updated. The old text is dated and still readable.
09
One machine, every way in.
Upload, AI agent, Teammate, Chrome clipper, Confluence import. Same checks. Same permissions.
Upload today
Upload PDF and Markdown today. More formats are coming.
Clipper and wiki
The Chrome clipper saves only the page you choose. Confluence imports as a copy with a link back.
Agents connect
Claude, Cursor, Codex, Devin, and any MCP-compatible agent connect today. They search first, then save what they settle.
Your wiki stays
Your wiki and search stay. Cluesora sits beside them.
Not available yet
Jira, email, and GitHub or other code hosts are planned. They are not available today.
- Upload
- Chrome clipper
- Confluence
- MCP agent
- Teammate
Same checks, same permissions, one map
Questions people ask.
Do I have to tag or file pages?
No. When content arrives it is read, summarised, connected, and checked. Nobody tags or files anything.
How can a question match if it uses different words?
The question is matched to what your organization decided and meant, not only to the words on a page.
How long until an upload is usable?
Small updates are usable within about a minute. Large PDFs take up to ten minutes.
Will an agent create a second copy of a doc?
Agents update the existing doc instead of forking a new one. Sources are never rewritten.
What if nobody has written the answer down?
Teammate finds the best-matched expert, asks them, and keeps the answer so the same question is not routed again.
Can I see why a result appeared?
Yes. Every result shows the exact line, its source, why it matched, and what it connects to.