−86%
Total cost
Average savings at 50+ users
Fixed per-user fee (BRL 89-149/month) replaces unpredictable per-token billing. BRL 423k (~USD 80k)/year saved in a real case.
OpenAI is the global LLM leader. MyDatAgent is the leader in private enterprise AI in Brazil. They're different products for different needs — and for most regulated Brazilian companies, the choice isn't as obvious as it seems.
Metrics from Brazilian companies that migrated to MDA in the last 12 months. Anonymized data from Enterprise customers.
Fixed per-user fee (BRL 89-149/month) replaces unpredictable per-token billing. BRL 423k (~USD 80k)/year saved in a real case.
Migration included in subscription. MDA team handles indexing, integration and training. 3-6 weeks, no additional cost.
All infrastructure (inference, logs, identity, audit) hosted in Brazil. ANPD audit passes without objections.
For regulated sector companies (finance, legal, health, government), these four pillars determine whether generative AI is viable or legal risk.
Per user/month, not per token. Finance approves AI budget with certainty. No surprise when the team increases RAG or multi-step agent usage.
100% Brazilian company. The US CLOUD Act doesn't apply to MDA. No data, metadata or telemetry leaves the country — auditable at any time by ANPD.
Parity with GPT-4 Turbo on PT-BR benchmarks (MMLU-PT, ENEM, BLUEX). ~120ms latency (vs ~600ms for OpenAI Brazil→Virginia). In multi-hop flows, it becomes seconds of waiting.
REST endpoint compatible with OpenAI SDK. Swap base_url and api_key — code stays the same. LangChain, LlamaIndex, Vercel AI SDK work without changes.
The 6 criteria that matter most to regulated Brazilian companies in 2026. No rhetoric — just verifiable data.
The MDA platform is model-agnostic. You can keep using GPT-4o, GPT-4.1 and o3 from OpenAI for specific cases where exact parity matters — just routed through our gateway, with PII redaction before it leaves, unified audit log, cost ceiling and automatic fallback.
Real scenarios: research team uses OpenAI's o1 for mathematical reasoning. Customer service uses MDA 2.1 (customer data never leaves). Marketing uses GPT-4o vision for creative analysis. All in the same gateway, same audit trail, same invoice.
CPF, RG, email and sensitive data automatically removed before any external call.
Every prompt, response, model and cost logged in one place — regardless of provider.
Quota per user, team and project. When it exceeds the limit, automatic fallback to cheaper model.
OpenAI down? MDA 2.1 takes over. High latency? Automatic routing to fastest model.
Set rules: sensitive data always MDA, mathematical reasoning o3, vision GPT-4o. By department, by tag, by user.
Model change is config update, not code change. Client app doesn't even notice.
Compiled from 200+ real questions asked by CIOs, CTOs, DPOs and Data Heads in MDA vs OpenAI evaluation cycles.
In Portuguese, yes — MDA 2.1 (32B parameters) matches GPT-4 Turbo on MMLU-PT, ENEM and BLUEX benchmarks. In mathematical reasoning and English code, it falls below GPT-4o but above GPT-3.5. For typical Brazilian enterprise use cases, the difference is imperceptible.
Yes. MDA's REST endpoint is compatible with the official OpenAI SDK. You just swap base_url and api_key. LangChain, LlamaIndex and Vercel AI SDK work without changes.
Average 60-86% savings for companies with 50+ active users. Real case: company with 80 users on GPT-4o was paying BRL 47k (~USD 8.9k)/month; after migrating to MDA's Growth plan, pays BRL 11.9k (~USD 2.3k)/month — BRL 423k (~USD 80k)/year in savings.
Technically yes, practically no. ANPD auditors have rejected OpenAI's DPA as sufficient legal basis for international transfer (Art. 33), because data remains subject to the US CLOUD Act. Regulated sector companies have been pressured to use Brazilian providers.
Yes to all. Function calling (with 31+ MCP Connectors ready), vision (image and document analysis), SSE streaming and batch processing. Full compatibility with OpenAI SDK.
~120ms average in Brazil. OpenAI has ~600ms (Brazil→Virginia round trip). In multi-step agent flows, the difference becomes seconds of waiting for the end user.
Not by default. Your data is used only at runtime (RAG) and stays isolated in your tenant. Fine-tuning is optional, executed in your tenant, and the resulting model is 100% yours — not shared with any other customer.
Book a demo at mydatagent.ai/contact. The MDA team handles the migration: data indexing, team training and integration with legacy systems. Average time 3-6 weeks. Enterprise customers get 30 days of parallelism with OpenAI key active.
The biggest barrier to migration isn't technical — it's organizational. Engineering teams worried about quality regression, finance holding up procurement, legal demanding proof of compliance.
MDA Consulting handles all three fronts: side-by-side benchmark with your real prompts, detailed financial plan with 12-month TCO, and LGPD dossier ready for internal audit. In 30 days you'll have a well-founded decision — without dragging a POC for six months.