Vectalk — Governed document and data intelligence
Ask your own documents anything. Inside your own walls.
VecTalk is ChatGPT, but on your contracts, policies, reports and records — deployed on your infrastructure or your cloud, with fixed rules deciding what is compliant and who may see what. Built for organisations that must follow the same rules up to the top 150, at a price written for everyone else.
3 vendor contracts expire this quarter with a penalty clause:
- Kaveri LogisticsExpires 30 Nov
Cl. 9.212% of the remaining contract value on early termination1
- Sahyadri Facility ServicesExpires 15 Dec
Cl. 12.1Two months of fees if the non-renewal notice is late2
- Narmada IT ServicesExpires 31 Dec
Cl. 7.45% of annual fees on exit before the term ends3
- 1Vendor_Agreement_Kaveri_2024.pdf, p.14, cl. 9.2
- 2Vendor_Agreement_Sahyadri_2023.pdf, p.9, cl. 12.1
- 3Services_Agreement_Narmada_2025.pdf, p.21, cl. 7.4
Three contracts. One clause. Nothing left the building.
A compliance head at a mid-size NBFC types:
“Which vendor contracts expire this quarter with a penalty clause?”
VecTalk returns the three contracts, the expiry dates and the penalty clause from each, with a link to the page it came from. She forwards it to the CEO. The documents never left her servers; the question never reached a foreign model.
1Answer · 3 contracts
2Source
9.2 Early termination
Should the Company terminate this Agreement before the Expiry Date for any reason other than a material breach by the Vendor, the Company shall pay the Vendor a penalty equal to twelve per cent (12%) of the remaining contract value, payable within thirty (30) days of termination.
Illustrative demo with fictional contracts — not connected to live data.
Questions your organisation cannot ask its data today
| Team | Asks VecTalk | Gets |
|---|---|---|
| Compliance & legal | “Which agreements carry an indemnity cap below ₹50 lakh?” | The list, each with the clause and page. |
| Finance | “What did we pay this vendor across all POs in FY 2025–26, and against which contract?” | The total, the POs and the governing contract. |
| Operations | “What is our SOP for a failed batch, and who signs off?” | The SOP section and the named approver. |
| HR & admin | “What does our leave policy say about encashment on exit?” | The paragraph, with the policy version and date. |
| Leadership | “Summarise last quarter's board papers on the Pune expansion.” | A sourced summary; nothing invented. |
Four things that are hard to copy
- 01
Your data stays with you
VecTalk runs on your servers, your private cloud, or — if you prefer — NeoPhoenix's own GPU infrastructure in India. Documents, embeddings and answers stay where you put them.
- 02
Fixed rules decide compliance — not the AI
Retention, residency and disclosure rules are written as policy and enforced by a deterministic rule engine. The model never decides what is allowed; it works inside what is allowed. Every determination is reproducible.
- 03
Access is decided before the AI answers
Roles, departments and field-level masking are applied before retrieval. A branch manager and a board member asking the same question get answers drawn from different documents, by design.
- 04
Every deployment faster and cheaper than the last
Pilot to production in 3 weeks, maximum. The pilot deployment is the production deployment.
Three ways to run it. All of them yours.
| Option | Where it runs | Commercials |
|---|---|---|
| Your infrastructure | Your data centre or server room | Licence + one-time setup + customisation and data preparation as needed |
| Your cloud | Your AWS, Azure or Indian-cloud account | Same as above |
| NeoPhoenix infrastructure | Our GPU servers in India, your tenant | Monthly, on credits and usage |
VecTalk runs on open-source models you control (the Mistral, Qwen and DeepSeek families, among others) — it replaces commercial AI, it does not wrap it. There is no API key to a foreign provider anywhere in the deployment.
The ₹6 lakh band, not the ₹6 crore one.
Pilots for mid-size organisations begin around ₹6 lakh a year. Enterprise deployments with multiple departments and agents run to the ₹85 lakh range. Every engagement is quoted on three lines — product licence, usage credits, and one-time setup — so a board sees exactly what recurs and what does not.
Get a quote on the briefing call- Pilots for mid-size organisations
- Begin around₹6 lakha year
- Enterprise deployments with multiple departments and agents
- Run to the₹85 lakhrange
Every quote has three lines
- 1Product licenceRecurs
- 2Usage creditsRecurs
- 3One-time setupOne-time
Frequently asked questions
It does what people use ChatGPT for — but on your own documents, inside your infrastructure, with your rules enforced before it answers. Nothing is sent to a commercial AI provider.
No. Documents, embeddings, prompts and answers stay on the servers you choose.
Open-source models that we fine-tune and deploy for you. You own the deployment; there is no dependency on a foreign API.
Those requirements are written as rules and enforced by a rule engine, not left to the model's judgement. Every decision is logged and reproducible for an auditor.
Yes — databases, APIs and document stores, with the same access rules applied to each.
Pilot to production in 3 weeks, maximum, counted from kickoff. The pilot is the production deployment.
Pilots start around ₹6 lakh a year; the briefing call gives you a three-line quote.
ISO 27001:2022 and ISO 42001:2023. A SOC attestation is in progress.
Opening the door is half the job.
Someone still has to do the work.
AI your board can say yes to.