Sales Coach · Security

Your data stays yours.

AI that reads your ERP has to earn trust with specifics, not adjectives. This page says exactly what runs where, what leaves your Odoo, what never does, and what we refuse to claim until it is true.

The model

Outbound only, aggregates only.

  • The module runs inside your Odoo.

    On your server, your hosting, your rules. All analysis of your sales history happens there, and the results are cached there too.

  • Traffic is outbound only.

    Your Odoo calls our service over HTTPS when it wants AI wording. The service has no inbound access to your database, ever.

  • Aggregates travel, history does not.

    A coaching request is a few kilobytes of computed figures about the current quotation. Your raw history is read locally and stays local.

The data flow

What a request carries, and what never leaves.

One compact JSON payload per coaching request, a few kilobytes. Disclosed in full, field by field, in the Sales Coach security docs.

A coaching request carries
  • The computed customer snapshot: cadence, typical order value, trend, risk class
  • The current order: its total, and each line’s product name, quantity, unit, price, discount
  • The detected signals: type, severity, estimated value, with internal record ids stripped
  • Your admin-written business profile text, locale, and currency name
  • The customer’s display name, only while the "send customer names" setting is on
  • A support fingerprint: a salted one-way hash of the database id (no personal data) and the installed module version
Never leaves your server
  • Raw order history
  • Your full product catalog and price lists
  • Line-by-line margins
  • Emails, documents, attachments
  • Anything about customers not on the current quotation

Transport is HTTPS only. The API key travels in a header on server-side calls from your Odoo; it is never exposed to browser code, and inside Odoo only Administrators can see it. The Ask tab works the same way: a chat question travels with the same compact, customer-scoped aggregates as a coaching request, and no new category of data leaves.

Commitments

Flat statements, not tiers.

  1. No training on your data

    We do not train models on your business data, and neither does our AI provider: text is generated by Anthropic’s Claude, which does not train on API data and processes requests transiently under a data processing agreement.

  2. Identifiable only if you choose

    Customer names in AI payloads are a setting. Turn it off and the service sees opaque references only. Any customer can be opted out of coaching entirely, and so can any product.

  3. Useful even with the AI off

    Disable AI enrichment and coaching continues deterministically inside your Odoo with built-in wording. If our service is unreachable, the panel degrades gracefully and your Odoo is never blocked.

The coaching service runs on Google Cloud in the EU (europe-west1, Belgium). Subprocessors: Google Cloud, Anthropic, and Razorpay for billing (we never see or store card data). Coaching requests and responses are processed transiently in memory and are not stored; the only response cache lives inside your own Odoo, with a six-hour default lifetime. Our data policy keeps usage telemetry to metadata only (timestamps, endpoint, latency, token counts, never request content), retained as raw events for twelve months and as monthly aggregates after that.

Where this is going

The perimeter is the roadmap.

Secure On-Prem AI is the design goal at the end of this page: models, data, and inference entirely inside your infrastructure, with nothing crossing out. It is in development, and the reference architecture is being defined with design partners. A concept, stated as one.

Concept diagram: Odoo and Nanti AI with local models operating inside the customer’s own infrastructure perimeter, with no connection crossing out to the public internet. SECURE ON-PREM AI — CONCEPT · IN DEVELOPMENT YOUR INFRASTRUCTURE Your Odoo Nanti AI local models · your hardware audit logs PUBLIC INTERNET no egress
The design goal for Secure On-Prem AI, in development now: models, data, and inference inside your perimeter — nothing crosses it. The reference architecture is being defined with design partners; this is a concept, not a shipped product.

Straight about status

What we do not claim.

  • No compliance certifications yet (no SOC 2, no ISO 27001). When an audit is passed, it will be listed here; until then we will not imply one with badges.
  • Fully on-premise AI is a design goal in development, defined with design partners. It is not a shipped product and we do not sell it as one.
  • No published uptime or customer numbers. We state what we can evidence.

This site walks the same talk: no cookies, no ad trackers, cookieless anonymized page counts, and the only personal data collected is an email you choose to give. Details on the privacy page. Ask us the hard questions: hello@nanti.ai