Weeks, not days
New engineers spend their first month reconstructing context that three people on the team already carry in their heads.
Now accepting early access
Your team already knows.
Your team already knows.
Now everyone can ask.
Klozon ingests your repos, Confluence pages, documents and notes — then answers any engineer's question with citations back to the exact source, and a full record of who contributed what.
Free to start. No credit card required. · already on the list.
app.klozon.co/#ingest
Ingest knowledgePayments Platform
4 channels
Git repositoryConfluenceFilesPaste
Clone & index
Cloned payments-api · 412 files
Converted 18 docs to Markdown
Indexed 1,284 chunks · attributed to Paramesh
AskAnswers grounded in your sources
How does refund settlement work?
Answer
Refunds post to the ledger immediately, but settle on the next batch run at 02:00 UTC. reconcile.py retries failures three times before routing to the manual queue.
docs/payments/refunds.md
reconcile.py
Confluence · Settlement runbook
Knowledge baseEvery source, and who added it
37 sources
SourceTypeAdded by
payments-api / READMEgitParamesh
Settlement runbookconfluenceMaria
PCI-scope.pdffileAlex
On-call handover notespasteJames
Prep planAlex Chen · day 3 of 7
82% ready
Read the service architecture
Set up a local environment
Trace a payment end to end
Ship a starter pull request
Quiz: settlement & refunds
Onboarding analyticsPayments Platform · 3 engineers ramping
+ New plan
Avg ramp time4.2d▼ 73%
Active onboarding12▲ 3
Knowledge coverage94%▲ 8%
Open questions0all answered
Active onboarding plans
AC
Alex ChenSenior Frontend Engineer
Day 3
MS
Maria SantosBackend Engineer
Day 1
JP
James ParkFull-Stack Engineer
Day 1
Ingests knowledge from
- Git repositories
- Confluence
- Word
- PowerPoint
- Excel
- Scanned docs (OCR)
- Pasted notes
The problem
Onboarding runs on tribal knowledge.
Onboarding runs on tribal knowledge.
And tribal knowledge does not scale.
The answer a new joiner needs is real — it is just scattered across a README nobody updated, a Confluence page from two reorgs ago, and a Slack thread three months deep. So they ask a senior engineer instead, and two people lose the afternoon.
Your best engineers, on tour-guide duty
Every unanswered question routes to whoever wrote the code. Their focus time becomes someone else's onboarding.
Documentation that decays silently
Nobody knows which page is stale until a new hire follows it into a dead end. There is no signal on what is out of date.
Answers you cannot verify
A generic chatbot will confidently invent an architecture. Without a citation back to a real source, nobody can trust it.
Where the time goes
Ramp-up is the most expensive week your team never measures.
Klozon makes it measurable — coverage, open questions, stale sources, and how fast each new joiner is actually getting there.
4.2dAverage ramp, instrumented
How it works
Three steps to a knowledge base that answers back.
Point it at what you already have. There is nothing to write from scratch.
Create a project
Each application or domain gets its own isolated workspace, so answers never leak across teams or blur two codebases together.
Ingest what exists
Paste a repository URL and it clones and indexes the README, docs and source. Add Confluence pages, drop in PDFs and Office files, or paste raw notes — everything is converted to Markdown and chunked for retrieval.
Ask, and follow the citation
Hybrid full-text and semantic retrieval finds the passages that matter. Every answer names the sources it came from, and every contribution is attributed in the audit log.
Features
Everything you need to onboard engineers at scale.
A single platform in place of scattered wikis, one-off Slack DMs, and the shadow-a-senior-dev-for-two-weeks approach.
Multi-channel ingestion
Git repositories, Confluence pages, file uploads and pasted content — one pipeline, one index, one place to look.
Document conversion & OCR
PDF, Word, PowerPoint and Excel are converted to compact Markdown before indexing. Scanned, image-only pages are read with on-device OCR.
Grounded, cited answers
Answers are built from passages that were actually ingested, and every one lists the sources behind it. If the knowledge is not there, it says so instead of guessing.
Provenance you can audit
A complete history of who ingested what and when, attributed to a real person — so you can trace any answer back to the contribution that produced it.
Onboarding analytics
Ramp speed, knowledge coverage, quiz scores, open questions and stale sources — measured from recorded activity, not self-reported status.
Isolated per project
Every project is its own workspace with its own sources, index and history, so two teams can share an instance without their answers bleeding together.
Pricing
Simple, transparent pricing.
Start on your own machine for free. Scale when the whole org wants in.
Starter
Try Klozon with your first project.
Free
No card required
- ✓ One project workspace
- ✓ Git, Confluence, file and paste ingestion
- ✓ Full-text and semantic retrieval
- ✓ Provenance and audit history
- ✓ Cited answers from your sources
Team
For engineering teams onboarding continuously.
$499/mo
Billed monthly
- ✓ Everything in Starter
- ✓ Approved enterprise LLM answers
- ✓ Role-based onboarding prep plans
- ✓ Onboarding analytics dashboard
- ✓ Scheduled Git and Confluence sync
- ✓ Contributor leaderboard
- ✓ Priority support
Enterprise
For large engineering organisations.
Custom
Talk to us
- ✓ Everything in Team
- ✓ SSO and SAML
- ✓ Per-project RBAC
- ✓ Audit log export
- ✓ Malware scanning on uploads
- ✓ Dedicated support and SLA
FAQ
Questions?
Questions?
We've got answers.
How is this different from pointing ChatGPT at our codebase?
A general assistant answers from what it inferred. Klozon answers only from passages that were actually ingested into your project, and names them. When the knowledge is missing, you get an honest gap instead of a plausible invention — and that gap shows up in analytics as something to go document.
What does it ingest?
Git repositories (it clones the repo and indexes the README, docs and source), Confluence pages, uploaded files, and pasted content. PDF, Word, PowerPoint and Excel are converted to Markdown first; scanned or image-only pages go through OCR locally.
What happens when the answer isn't in our knowledge base?
It tells you, rather than filling the gap with something plausible. Answers are assembled only from passages that were actually ingested, so a missing answer is a real signal — and it surfaces in analytics as a documentation gap worth closing.
Where are model credentials kept?
Server-side only. The proxy holds the key and the browser never sees it; a health endpoint reports whether a model is configured without exposing anything. Nothing sensitive is stored in client code, browser storage, or the repository.
Can I keep several applications separate?
Yes — that is the default. Every project is an isolated workspace with its own sources, its own retrieval index and its own history, so two teams can share an instance without their answers bleeding together.
Who gets credit for contributions?
Every ingestion is attributed to the person who performed it and recorded in the audit log, with a contributor leaderboard on the project overview. Sharing knowledge becomes visible work rather than invisible favours.
What do I need to set up?
Nothing to install and no build step. Create a project in the browser, paste a repository URL or drop in some documents, and ask your first question — the whole loop takes about five minutes.
What is on the roadmap?
SSO and SAML, per-project role-based access, scheduled re-sync for Git and Confluence so sources never drift, and malware scanning on uploads. Early-access teams help set the order.
Stop losing weeks to knowledge that already exists.
Create a project, point it at one repository, and ask it a question. The whole loop takes about five minutes.