Now liveMade for teams that work with data

Your company's brain,
on demand.

Upload files, connect Gmail, Calendar, Notion, your CRM, then ask anything, by chat or voice. A team of expert agents routes, reasons, and answers from your data with citations you can click and numbers computed with real SQL.

Book a demo →orCreate workspace

30-minute walkthrough with your data · Or start free, no credit card

12+
Connectors & databases
6
Expert agent domains
100%
Answers cited to source
0
Invented numbers
What it looks like
app.vectropic.ai/chat
Vectropic
Chats
Files
Connectors
Studio
Recent
Q3 revenue by region
Vendor contract terms
Top P1 tickets
What was our total revenue by region in 2025?
Vectropic.ai

2025 revenue by region: India $6.2M, EMEA $2.6MSQL. EMEA grew 42% YoY, driven by the Heineken rollout1

SELECT region, SUM(revenue_usd) FROM sales_2025 GROUP BY region
1aurora_sales_2025.csv · rows 214–262
Why we built this

Most AI assistants predict an answer and hope it's right. Vectropic retrieves, computes, and reasons, then shows its work. If the answer isn't in your data, it says so. If it is, you can click straight to the line it came from.

Capabilities
01

Answers with receipts

Every reply cites its source. Click a citation to see the exact passage highlighted, or the exact spreadsheet rows.

02

Exact numbers

Spreadsheets and PDF tables become real SQL tables. Revenue by region is computed, not guessed.

03

Experts that reason

Every question is routed to a professional: financial analyst, technical writer, legal, HR, sales, or data. Hard questions get broken down, researched, and reasoned through step by step.

04

All your tools

Gmail, Google Calendar, Drive, Notion, Slack, Confluence, Salesforce, HubSpot, all synced in the background, incrementally, into one searchable knowledge base.

05

Build your own agents

The Studio: design custom agents and multi-step flows in plain English, run them on a durable queue, and audit every run.

06

Watch it think

Flip from Chat to Visual and see retrieval, routing, and reasoning play out live, or just ask by voice and listen.

Grounded answers

Every claim traces back to a line in your data.

Documents are chunked, embedded, and retrieved semantically, then graded. Weak evidence gets re-routed to a better source instead of papered over with confident prose.

  • Clickable citations. Open the exact passage, highlighted, or the exact spreadsheet rows.
  • Honest gaps. When your data doesn't cover it, the answer says so and points to what it does cover.
  • Any format. PDFs, Word, Excel, CSV, emails, wiki pages, CRM records.
Answer tracechat · sources
Question"What's our parental leave policy in Germany?"
Retrievedpeople-handbook.pdf · §4.2 Parental leave · score 0.91
Gradedrelevant ✓ · sufficient ✓
Answer14 weeks fully paid, extendable to 12 months unpaid [1]
Source [1]people-handbook.pdf · page 12, lines 8–19
Exact numbers

Figures are computed with SQL, never predicted.

Every spreadsheet and PDF table you add becomes a real, typed database table. When you ask a quantitative question, the pipeline writes a guarded, read-only query, runs it, and shows it to you next to the answer.

  • Inspectable. The SQL chip on every computed answer opens the query that produced it.
  • Guarded. Read-only, injection-checked, row-capped. It can't touch your data, only read it.
  • Comparative. Group-bys, trends, top-Ks, and cross-site comparisons, straight from plain English.
Text → SQLread-only
Question"Compare average machine uptime between our sites"
SELECT site, AVG(uptime_pct) AS avg_uptime FROM machine_telemetry GROUP BY site ORDER BY avg_uptime DESC
siteavg_uptime
Berlin96.24
Pune95.91
Layered reasoning

A team of experts, not one generic bot.

Every question passes through a three-layer pipeline. A router classifies the domain and difficulty. A specialist (financial analyst, technical writer, legal, HR, sales, or data) answers in the right voice with the right rigor. Hard questions get decomposed into sub-questions, researched in parallel, and synthesized.

  • One classification drives everything. No duct-taped chains of prompts.
  • Domain contracts. The financial expert shows derivations; the technical expert gives runnable curl commands.
  • Visible thinking. The plan and each reasoning step stream into the UI as they happen.
Agent pipelinelive steps
L1 Routedomain: financial · complexity: multi-step
L3 Plan3 sub-questions → revenue, one-offs, YoY basis
Step 1/3Q4 revenue by region · SQL over sales_2025 ✓
Step 2/3one-off items · search finance-notes.pdf ✓
Step 3/3YoY basis · comparing against 2024 table…
L2 Expertcomposing as Financial analyst · cited
The Studio

Describe an agent. Get a working one.

Tell the Studio what you need, like "classify inbound emails, extract the order number, and log it", and it designs the agent: persona, tools, and a multi-step flow. Runs execute on a durable queue that survives crashes and retries failures.

  • Plain-English building. The copilot drafts the agent; you refine it in chat.
  • Real workflow steps. Classify, extract, query data, search docs, call APIs, branch on conditions.
  • Full history. Every run and every change is kept, step by step, with request-level audit.
Run historyinvoice-triage-agent
run_0292done · 2.1s · classify → extract → write_record ✓
run_0291done · 1.8s · trigger: chat
run_0290done · 2.4s · trigger: manual
auditPUT /agents/… → 200 · "raised limit to 50" · by maya@…
auditPOST /agents/…/run → 201 · body logged, keys redacted
Connectors

Everything your team already uses, in one place.

Connect once. Syncs run in the background, pick up only what changed, and prune what was removed, so your knowledge base stays current without anyone thinking about it.

Gmail
Recent mail, scoped and incremental, with promotions and spam filtered out.
Google Calendar
Meetings as clean context: recurring series are collapsed, not 200 duplicates.
Google Drive
Docs, Sheets, and Slides, kept in sync as they change.
Notion
Every page shared with the integration, fully paginated.
Slack
The decisions buried in channels become searchable answers.
Confluence
Your wiki, finally answerable.
Salesforce
Accounts, deals, and pipeline, queryable in plain English.
HubSpot
Contacts and deals alongside the rest of your knowledge.
PostgreSQL
Point at a database; ask questions in plain English.
MySQL
Same for MySQL: read-only, guarded queries.
MongoDB
Collections included in the same question surface.
Meeting bots
Google Meet, Zoom, and Teams transcripts.
Coming soon
Use cases

One assistant. Every team's questions.

The same workspace answers finance, sales, HR, legal, engineering, and ops, each in the right expert's voice, each grounded in your data.

Finance

Close the books faster

What was our total revenue by region in 2025?
Which invoices over $10k are unpaid past 30 days?
Compare average spend per vendor this quarter vs last.
Sales

Know the pipeline cold

Which deals slip most often, and why?
Summarize every touchpoint with Heineken this quarter.
Who are our top 10 accounts by expansion revenue?
HR & People

Answer policy questions instantly

How many parental leave days do employees in Germany get?
What's the process for a role change?
Draft an onboarding checklist from our handbook.
Legal

Find the clause, cite the page

What are our termination obligations in the Acme MSA?
Which contracts renew in the next 90 days?
Does any vendor agreement allow data sub-processing?
Support & Eng

Docs that answer back

How do I call the invoices API? Give me a curl example.
What changed in the v2 webhook payload?
Which P1 tickets mention the export feature?
Data & Ops

Ask instead of querying

Compare average machine uptime between our sites.
Which 3 customers received the most units?
Show monthly ticket volume as a trend.
Process

Three steps.
That's it.

01 →
Create a workspace

Sign up, invite your team. Every org's data is fully isolated.

02 →
Add your data

Drop in PDFs and spreadsheets, or connect Gmail, Calendar, Notion, your CRM, synced in the background.

03 →
Ask anything

Chat or talk. Expert agents reason over your data and answer with citations and exact figures.

Under the hood

Built like infrastructure, not a demo.

The parts you never see are the reason the answers hold up.

Retrieval

Hybrid search over vectors

Documents are chunked and embedded into pgvector; retrieval blends semantic similarity with grading, so weak matches are re-routed instead of hallucinated over.

Agents

A LangGraph pipeline, end to end

Routing, retrieval, SQL, reflection, and multi-step reasoning are nodes in one graph: inspectable, testable, and streamed to the UI step by step.

Compute

Real SQL, not arithmetic by vibes

Spreadsheets and tables become actual database tables. Quantitative questions compile to guarded, read-only SQL you can inspect in the answer.

Sync

Incremental, background, resumable

Connector syncs return instantly and run in the background. Already-ingested content is skipped by checksum; removed content is pruned.

Execution

A durable agent queue

Custom agent runs go through a persistent queue with retries, heartbeats, and rescue. A crashed worker never loses a run.

Accountability

An audit trail for everything

Every mutating action is logged with actor, path, and a redacted copy of the request. Agent and flow runs keep full step-by-step history.

Security & control

Your data stays yours.
And you can prove it.

→
Tenant isolation

Every workspace is a separate tenant. Files, embeddings, tables, and chats are scoped to your organization on every single request.

→
Role-based access

Admins, members, and per-connector credentials. Read-only, guarded access to any database you connect.

→
Full audit trail

Every change is recorded (who, what, when) with sensitive fields redacted before anything is stored. No training on your data, ever.

FAQ

Fair questions.

What is Vectropic.ai, in one sentence?
A workspace where your company's files, tools, and databases become one assistant that answers with citations, exact numbers, and expert-level reasoning.
How is this different from ChatGPT with uploads?
Three ways. Grounding: every answer cites the exact passage or spreadsheet rows it came from. Numbers: tables become real SQL tables, so figures are computed, not predicted. Structure: questions are routed through a layered agent pipeline where a domain expert answers, and hard questions are decomposed and reasoned through step by step.
Is my data isolated from other companies?
Yes. Every workspace is a separate tenant. Files, embeddings, tables, and chats are scoped to your organization at the database level, and access is enforced on every request.
Does it train on my data?
No. Your data is used only to answer your team's questions. It is never used to train models.
What can I connect?
Files (PDF, Word, Excel, CSV, and more) plus Gmail, Google Calendar, Google Drive, Notion, Slack, Confluence, Salesforce, HubSpot, and direct database connections for PostgreSQL, MySQL, and MongoDB. Meeting transcripts from Google Meet, Zoom, and Teams are coming soon.
What are custom agents and flows?
In the Studio you describe an agent in plain English, like "classify inbound emails and extract the order number", and it builds the agent, its tools, and a multi-step flow. Runs execute on a durable queue, and every run keeps a full step-by-step history you can audit.
Can non-technical people use it?
That's the point. Ask in plain language, by chat or voice. The SQL, retrieval, and reasoning happen underneath: visible if you want them, invisible if you don't.
How do I get access?
Book a demo and we'll set your workspace up together on a 30-minute call. Prefer to explore on your own? Create a workspace at app.vectropic.in, free to start, no credit card.
Get started

Ready when you are.
Free to start.

Book a demo →orCreate workspace

30-minute walkthrough with your data · Or start free, no credit card