NewEvery answer now highlights the exact passage→

Ask your company what it already knows.

Flintyst connects Slack, Drive, Confluence, Jira and dozens more, then answers questions in plain language with every claim cited. It only uses what each person is already allowed to see, and it keeps up as your tools change.

Live in a weekCited answersPermission-awareNo per-seat fees
flintyst.techyst.net / acme / askIndex current
Permission-aware · 12 sources connected

Answers stream here, with a numbered citation on every claim.

Try asking
SourcesWaiting for a question

Pick a question, press / to type your own, and hover a citation to see the exact passage.

Reads from the tools your company already writes in

SlackGoogle DriveConfluenceJiraGitHubNotionGitLabGoogle SitesMicrosoft Teams
SalesforceZendeskSharePointGmailWeb crawlFile uploadsDropboxLinearFreshdesk

The problem

The answer exists. Finding it is the job.

Decisions live in threads, specs in wikis, context in tickets and the final word in someone's inbox. Every question becomes a scavenger hunt across six tools and two colleagues.

  • One question, every source. Flintyst searches all of them at once, ranked by what actually answers it.
  • The current version wins. Recency, authority and feedback push stale pages down.
  • Nobody gets interrupted. The answer comes from the record, not from pinging whoever wrote it.
Without Flintyst46 min

Search Slack for “EU residency”212 results

Three pages called “Data residency”Which is current?

DPA addendum v2 (final) (1).docxOr is it v3?

“Does anyone know…?” in #eng-platformWaiting

“Check the ADR, I think”17 min later

Find ADR-031, unsure it's the latest46 min gone

With Flintyst4.2 s

What did we decide about EU data residency?

Pinned to eu-central-1 for enterprise tenants 1, including indexes and backups 2, with the dedicated cluster due 28 November 3.

4 sources · cited

Connected sources

Every tool you use, one index that keeps up.

Connectors sync documents, messages, tickets and code into a single index, along with who is allowed to see each one. Click any source to see what it pulls and how its permissions come across.

1,231,167documents indexed

What's inside

More than a chatbot on top of a search box.

Retrieval, ranking, permissions and citations are the product. The chat window is just where you meet them.

Answers with receipts

Every sentence points to the passage it came from. Hover a number and the quote lights up in its source.

The on-call rota changes on Monday 1 and swaps are made in PagerDuty, not Slack 2. Marco covers the first week 3.

1On-call policyRota changes every Monday at 10:00.
2#sre announcementPlease make swaps in PagerDuty.
3OPS-77 Rota Q4Week 1: Marco R.

Hybrid search

Keywords for exact names and IDs. Vectors for meaning. Both, reranked.

keywordmeaning

Always current

Incremental syncs every few minutes, so this morning's decision is already searchable.

Synced 2 min ago

Search mode, too

Sometimes you want the document, not a summary. Filter by source, author and date.

Source: ConfluenceLast 90 days

Q4 pricing change: FAQ0.94

Pricing deck v120.88

#pricing decision thread0.81

Right inside Slack

Mention the bot in any channel. The answer lands in the thread, cited.

JTJonah @flintyst what's the VPN for the Lisbon office?

Flintyst Use the eu-south profile in the VPN client 1. Setup steps are in the IT guide 2.

Gets better with feedback

Thumbs up or down on an answer adjusts how its sources rank next time, and admins see which questions went unanswered.

ADR-031: EU data residency▲ boosted

Data residency FAQ (2023)▼ demoted

DPA addendum v3▲ boosted

Old pricing deck v9▼ demoted

An API for everything

Put cited answers inside your intranet, your support desk or your own tools.

# Illustrative request
curl -X POST https://flintyst.techyst.net/api/ask \
  -H "Authorization: Bearer $FLINTYST_KEY" \
  -d '{"question": "Who owns billing?"}'
→ { "answer": "The payments team…", "citations": [3] }

How an answer is made

Six steps, under a second.

Nothing mysterious happens between the question and the answer. Step through it, and watch the reranker throw out a page that only looked relevant.

Step 1 of 6 · 4 ms

Understand the question

The question is parsed for intent, entities and time, then expanded with the terms your company actually uses.

End to end 0.72 s Illustrative

“What did we decide about EU data residency for enterprise customers?”

intentdecisiontopicEU data residencyaudienceenterpriseasked byPriya · 3 groups

Expanded with +gdpr +eu-central-1 +data location

Permission-aware

Same question. Different people, different answers.

Flintyst checks each source's own access rules before retrieval, so the model never reads what the person asking couldn't open themselves. Switch who's signed in and ask again.

  • Synced, not re-typed. Channel membership, sharing lists and project roles come across with the documents.
  • Filtered before the model. Restricted passages are never retrieved, so they can't leak into a summary.
  • Groups and document sets handle anything a source can't describe on its own.
flintyst · signed in as Priya NairPermission-aware
PNPriya Nair · Platform engineerengineeringacct-northwind

Where are we on the Northwind renewal, and what's blocking it?

6 candidates matched3 allowed for Priya3 filtered before the model saw them

Cited · 3

1#acct-northwindPrivate channel · 9 members
2SEC-231 SCIM provisioningJira · Security · In progress
3Enterprise SSO & SCIM guideConfluence · Public docs space

3 sources hidden: no access

Hidden source.Salesforce sharing rules: Sales profile only
Hidden source.Doc is shared with the sales group only
Hidden source.Mailbox owner and delegates only

Shown here for the demo. In Flintyst, sources you can't open never appear at all.

Custom assistants

One brain, many specialists.

Give each team an assistant with its own instructions, its own document sets and its own audience. Pick one to see it answer.

Across the company

Every department has a question it keeps asking.

The questions change by team. The shape of the answer doesn't: direct, current and cited.

Stop re-deriving decisions from old threads. Find the ADR, the PR and the incident that explain why the code looks the way it does.

Sources that matter most
GitHubJiraConfluenceSlack
First assistant to set upOn-call Helper

Why did we move session storage off Redis?

Memory pressure during the March incident; ADR-024 chose Postgres with a TTL index. 1

ConfluenceJira

Who owns the billing service now?

The payments team since the Q2 re-org; Marco is the tech lead. 1

ConfluenceSlack

Is there a runbook for rotating the TLS certs?

Yes, in SRE runbooks, and it was updated after last month's renewal. 1

ConfluenceGitHub

The job

Answer model

Which runbook covers a Postgres failover?

Asked in On-call Helper · answered by Fast
Answer qualityGood
Speed0.6 s
Cost per answer1× Fast
Typical answer0.6 s
Context window32k tokens
Relative cost1×

Best fit for this job. Short, factual answers from one or two passages. Speed matters more than depth here, and it is the cheapest model per answer.

Your assistants

Every model answers only from retrieved passages, with citations. Figures are illustrative.

Choose the model that fits

The right model for every job.

A quick lookup doesn't need the biggest model, and a contract review shouldn't get the smallest. Pick a job, try a model, and see the trade-off in quality, speed and cost.

  • A different model per assistant. A fast one for lookups, a deeper one for analysis, a long-context one for contracts.
  • Switch whenever you like. Changing models never means re-indexing a single document.
  • The same receipts either way. Every model answers only from retrieved passages, with citations and permission checks.

From kickoff to company brain

Three weeks from first source to everyone asking.

A typical rollout, run with our team alongside yours. Scroll through it.

Week 1

Set up the workspace

We create your Flintyst workspace, connect single sign-on through your identity provider, provision people and groups, and agree on admin roles with you.

Workspace setup · Acme

Workspaceacme · created

Sign-inOkta · SAML

People and groupsSCIM · 1,240 people

Admin roles4 admins · 2 auditors

Answer modelsBalanced · per assistant

Week 1

Connect the first sources

Start with the three tools people search most. Each connector is authorised by its own admin, with read-only scopes.

Connectors

SlackConnectedOAuth · read-only

ConfluenceConnectedAPI token · 14 spaces

Google DriveConnectedService account · domain-wide

JiraConnectedOAuth · 9 projects

SalesforcePendingWaiting for admin consent

Week 2

Build the first full index

History is indexed once, then kept current with incremental syncs. Permissions come across with every document.

First full indexETA 3h 10m

Slack412k docs
Google Drive152k docs
Jira96k docs
Confluence38k docs

Permissions indexed alongside every document

Week 2

Shape assistants and access

Map groups to document sets and give each team an assistant with instructions that fit how it works.

Groups → document sets

EngineeringRunbooksADRsRepos
SalesDealsSecurity docs
PeopleHandbooksBenefits
EveryoneCompany wikiAnnouncements
Week 3

Roll out, then learn

Open it to the company. Feedback and unanswered questions show which pages to fix and which sources to add next.

Questions per day · first monthIllustrative

Top unanswered this week

Where is the brand font licence?add a source

What's the SLA for Enterprise support?add a source

The difference

Six search boxes, or one question.

Every tool has its own search. None of them knows what the others say, which version is current, or who is allowed to see what.

Searching tool by toolFlintyst
Where you lookSix search boxes, one at a timeOne question across every source
What comes backA list of links to open and skimA direct answer, every claim cited
Which version is currentWhoever edited last, or nobody knowsRanked by recency, authority and feedback
PermissionsChecked by hand, or not at allSynced from each source, checked on every query
Time to an answerMost of a morning, plus a colleagueA few seconds
Fit for each teamOne generic search for everyoneAssistants with their own sources and model
PricingPer seat, every monthNot per seat. Add the whole company.

Control

Built for companies that answer for their data.

Single sign-on

OIDC or SAML through your identity provider. People are who your directory says they are.

Permission sync

Access rules come across from each source with the documents, and are checked again on every query.

Query audit

Who asked what, which assistant answered and which sources it used, kept as long as your policy says.

Encrypted throughout

Documents, embeddings and questions are encrypted in transit and at rest, and connector tokens are stored encrypted.

Never training data

Your documents and questions are not used to train anyone's model, whichever model answers.

Backups, uptime and export

Daily backups verified by restoring them, monitored uptime, and a full export of settings and history whenever you ask.

Query auditacme workspace · liveIllustrative

priya.nOn-call Helper3 sources

diego.mSales Engineer4 sources

sam.kGeneral1 source · 5 filtered

lena.wHR Policies2 sources

marco.rGeneralno match

aisha.kOnboarding Buddy2 sources

jonah.tSlack bot3 sources

FAQ

Questions, answered.

If your question isn't here, ask us. A person who works on Flintyst every day will answer, with sources.

How long does it take to get started?

Most teams ask their first real questions within a week. We set up your workspace and single sign-on, connect the first sources with you, and the full index builds in the background. A company-wide rollout typically takes about three weeks.

Which tools can Flintyst connect to?

Slack, Google Drive, Confluence, Jira, GitHub, Notion, Salesforce, Zendesk, SharePoint, Gmail, Microsoft Teams, Linear, Dropbox and more, plus web crawls and file uploads. Each connector is authorised by that tool's own admin, with read-only scopes.

Which model answers questions? Can we choose?

Yes. Use a fast model for quick lookups, a deeper one for analysis or a long-context model for contracts and reports, and set a different one for each assistant. You can switch later without re-indexing, and every model answers only from retrieved passages, with citations.

How do permissions work?

Where a source exposes its access rules, Flintyst syncs them with the documents and checks them on every query. Someone who can't open a file in Drive won't get an answer built from it.

For everything else, you scope access with groups and document sets.

Does it make things up?

Answers are generated only from retrieved passages, and every claim carries a citation you can open. When the index doesn't cover a question, Flintyst says so instead of guessing. You can still judge an answer, and now you can check it in one click.

How fresh is the index?

Most connectors sync incrementally every few minutes, and some pick up changes as they happen. Each source shows when it last synced, and admins can trigger a sync at any time.

Is our data used to train models?

No. Your documents and questions are never used to train anyone's model, whichever model you choose. Everything is encrypted in transit and at rest, and every query is recorded in an audit log your admins can review.

How does pricing work?

Pricing isn't per seat, so giving the whole company access costs the same as giving it to one team. Ask us for a quote.

Can people use it from Slack?

Yes. Mention the bot in a channel or send it a direct message and it answers in the thread, with the same citations and the same permission checks as the web app.

Can we put Flintyst answers inside our own tools?

Yes. The API returns the same cited answers as the web app, so you can add them to an intranet, a support desk or an internal tool. Every request is checked against the permissions of the person asking.

Stop searching. Start asking.

Connect your first three sources, ask the question your team has been chasing all week, and check the citations yourself.