Home / Platform
000  /  Platform Overview

The control plane for regulated AI.

One layer between every AI application and every compute environment. frIdA routes each workload by residency, latency, cost and policy. It enforces compliance at execution and logs every decision for audit.

Cloud, hybrid, on-premise or air-gapped, under one policy framework. Forged in LatAm, not limited to it.

002 / Every Decision On The Record

One workload, from prompt to proof.

We don't watch the model and write a report afterward. frIdA sits in the path. Every workload gets a verdict, a cited rule, and a sealed record before it ever reaches production. Here is a single credit request, the way the platform sees it.

Decision #MX-CR-48217 · BANK_MX · frIdA runtime
Logged · immutable
// 00Capture
A credit-origination copilot submits a customer-facing request: generate the adverse-action notice for a declined auto loan. The prompt carries the applicant's data and the institution's identity. workload: credit.adverse_action_notice tenant: BANK_MX contains: applicant PII · decision rationale
// 01Evaluate
frIdA reads the policy that applies to this tenant and this data class before anything executes. Residency, applicable frameworks, and PII handling are resolved at execution time, not assumed from a contract. residency: MX.sovereign frameworks: CNBV · CONDUSEF pii_scan: enabled
// 02Decide
The policy resolves to a verdict and a route. The applicable rule is named on the decision itself, so the reasoning is legible to a risk committee, not buried in a log nobody reads.
Allow · with PII redaction
Rule cited: CNBV CUB · transparency duties under CONDUSEF
// 03Deliver
The notice runs inside the sovereign environment the policy demanded. It comes back in Spanish, with applicant PII masked and the required disclosures present. No data left the jurisdiction to produce it. route: SOVEREIGN · MX output: es-MX · PII masked · disclosures ✓
// 04Log
The whole path is sealed into one immutable record: who asked, which policy applied, which rule fired, where it ran, which model answered. When the regulator asks what happened, you hand over the record. You don't run a query and hope. record: decision_id · timestamp · policy_v · route · model export: regulator-ready
003 / The Platform

frIdA decides. Studio is where you build and run it.

Deployment modes and the security posture go deeper on their own pages: Deploy Anywhere → · Security & Compliance →. Both are what frIdA does, not separate products.

004 / How frIdA Routes

Four signals decide where a workload runs.

Every workload is scored against the same four signals before it executes. The route is not a default someone set once. It is recomputed per request, and the reasoning is logged.

Signal
What frIdA reads
Effect on the route
Residency
The jurisdiction the data belongs to and the law that binds it.
Pins execution to a permitted environment. A workload that would cross a border does not route.
Latency
How fast the workload has to answer to be useful in the loop it serves.
Prefers the closest compliant environment that meets the deadline.
Cost
The price of each eligible model and environment for this class of work.
Picks the cheapest option that still clears residency and latency.
Policy
The institution's own rules: data classes, approved models, access scope.
Overrides everything. Policy is the ceiling, never the thing optimized away.
005 / Enablement

Knowledge transfer is a product, not a service.

Every engagement ships with a structured enablement program. Run inside your organization. Your team gets autonomous on what we deployed.

001 / CURRICULUM

Structured, not improvised.

Every module maps to a specific capability you deployed. Your team learns to operate, extend, and audit the applications without us. Autonomy is the exit criterion.

002 / CADENCE

The frontier moves every quarter.

New module each cycle. Enablement closes in weeks, not the months enterprise software procurement takes. Each module opens the next use case.

003 / DEPENDENCY

The relationship has to be exitable.

If the only reason you still need Saptiva AI is that your team can't operate the platform, we haven't delivered. Enablement is how dependency becomes capability.

004 / OUTCOMES

Capability is the KPI.

ACMES's team audits their own policy files. Ibero's faculty extends their own Studio apps. The exit criterion isn't training completed. It's capability owned.

006 / What The Platform Is Built On

Four principles we will not negotiate.

Every architectural decision traces back to one of these. Our customers cannot afford for them to be otherwise.

001 / RESIDENCY

Data stays where law requires it.

Enforced at execution time by frIdA, not by contract. A workload that violates your residency rule doesn't route. The platform refuses to execute. Residency is architectural, not administrative.

002 / PORTABILITY

Zero lock-in by architecture.

You choose the model, the cloud, the deployment mode. Swap the LLM and frIdA routes around it. Change the cloud and the same policy applies. If we can only keep you by trapping you, we haven't earned the relationship.

003 / AUDITABILITY

Every decision is logged.

Every routing decision, every policy evaluation, every execution produces an immutable record: readable, exportable, regulator-ready. If something goes wrong, you can prove what the platform did and why.

004 / PRODUCTION FOCUS

Pilots are how AI fails here.

Production from day one. Our embedded engineers land inside your team, ship the first use case in two weeks, and stay until it runs. We optimize for what still runs eighteen months later.

007 / Get In Touch

Technical deep dive on any layer.

In production at Rappi, Banco Invex, MultiMoney, Universidad Iberoamericana, and ACMES. KAL, Mexico's national LLM, runs on the platform under Mexican jurisdiction.

If you're evaluating AI infrastructure for a regulated institution and want to go below the overview, our engineering team responds within 48 hours. Not a sales sequence.

Talk to an engineer See production deployments →