10 Trillion Tokens, One Decision: The Router Behind Brazil’s Largest AI Platform

Highlights

• Adapta achieved over 70 million in recurring revenue while remaining profitable without outside funding or subsidizing tokens.

• Adapta launched in 2023 and reached 25,000 customers within six months.

• The Adapta platform uses a custom router to select the optimal model for each prompt based on an evolving golden dataset, reducing costs while maintaining frontier model quality.

• An agriculture consultancy customer doubled their service capacity and saved over $200,000 using a custom agentic CRM built on Adapta's platform.

• Adapta deploys remote engineers to help 60 companies at a time navigate the missing middle between product-led growth and high-touch enterprise contracts.

Timeline

1. 00:00:00 - 02:40:00

   Adapta's Origins and Initial Growth

   Adapta emerged as a Brazilian bootstrap company focused on teaching users how to extract practical value from AI. By combining specialized knowledge work tools with education, the platform gained rapid traction and secured major sponsors like Microsoft, AWS, and Google for its initial events.

   • Adapta launched in 2023 to address the gap between what AI models could do and how users could apply them.
   • The platform bundled knowledge work features such as context building, memory, and prompt builders with training and education-based marketing.
   • The first event in early 2024 drew 4,000 attendees, and customer count reached 25,000 within six months of launch.
2. 02:40:00 - 08:00:00

   Model-Agnostic Routing and Vercel Integration

   As model options expanded beyond GPT, managing individual API rules and provider negotiations became unsustainable. Adopting an AI gateway centralized observability and infrastructure management. The custom router layer reduced operational costs while maintaining frontier model performance based on historical performance data.

   • The proliferation of competitive models created a new challenge regarding model selection and maintenance complexity.
   • Adapta built Adapta1 and integrated an AI gateway to manage multiple model providers and eliminate disparate dashboard rules.
   • The internal router evaluates prompts against an evolving golden dataset of user feedback to automatically select the most cost-effective and capable model.
3. 08:00:00 - 11:11:00

   Continuous Learning and Agentic Workflows

   Beyond routing external APIs, the platform incorporates a self-improving proprietary model that resolves a large volume of user prompts natively. The Maestro platform expands on this by enabling companies to execute heavy knowledge work through multi-agent systems.

   • Adapta developed its own proprietary model trained on open-source foundations to handle routine user prompts alongside external models.
   • The proprietary model learns continuously from every user interaction while retaining prior knowledge.
   • The Maestro product introduced full-agent capabilities for complex knowledge work and company-wide agent deployment.
4. 11:11:00 - 15:42:00

   Solving the Missing Middle with Remote Deployed Engineers

   Mid-sized service businesses face high complexity that simple tools cannot solve, yet they lack the budget for traditional forward-deployed enterprise engineers. Adapta bridges this gap by deploying remote engineers who help multiple companies build bespoke agentic workflows, exemplified by a client who automated hundreds of spreadsheets into an agentic CRM.

   • Traditional product-led growth is insufficient for mid-sized businesses, while high-touch enterprise deployment engineers are too resource-heavy.
   • Adapta uses remote deployed engineers to support 60 mid-market companies simultaneously instead of assigning dedicated on-site personnel.
   • An agriculture consultancy doubled its service capacity and saved over $200,000 by building a custom agentic CRM on the platform.