• 

  Opti.

Optim

Inspiration

Enterprise AI spending has become a financial black box. Companies receive massive month-end invoices with no way to trace costs to specific teams, products, or clients. Engineers default to the most expensive models for routine tasks (when alternatives 30-50x cheaper would do the job), and SaaS companies can't calculate true unit economics on AI-powered features. We saw three failures playing out at once: bill shock, systematic overspending, and broken margins — and no neutral layer to fix them.

What it does

Optim is a financial governance layer that sits between companies and their AI providers. It attributes every cent of AI spend to the exact request that generated it using a three-variable cost model (which model, output depth, prompt count), then autonomously routes each request to the most cost-effective capable model in real time. The result: a real-time ledger of AI spend, model arbitrage across providers, OpEx forecasting, and true unit economics per feature, request, or client.

How we built it

We built a provider-agnostic routing infrastructure hosted in Europe with RGPD compliance designed into the architecture from day one. The core is a three-dimensional cost model that produces a "cost fingerprint" for every individual request — the atomic level of cost attribution. On top of that sits an agentic optimization layer where AI agents analyze usage patterns in real time and either recommend or autonomously execute routing decisions, rather than logging data for batch human review.

Challenges we ran into

• Operating in a space with established US incumbents (Helicone, Langfuse, Portkey, Datadog LLM Observability) and a narrowing European window
• Convincing mid-sized engineering teams to choose us over a DIY "logger plus Grafana dashboard"
• Staying ahead of native cost dashboards from OpenAI, Anthropic, and Google as the observability layer commoditizes
• Translating strong founder networks into externally visible market validation

Accomplishments that we're proud of

Building true provider agnosticism — no commercial relationship with any AI vendor biases our routing. Achieving request-level cost granularity, which is the technical ceiling for attribution accuracy. Designing RGPD compliance into the architecture rather than retrofitting it. And shifting the category from passive observability to autonomous optimization — a cost reduction system, not a cost dashboard.

What we learned

Visibility isn't the value — control is. Companies don't need another dashboard; they need the ability to act on cost data without human intervention at the request level. We also learned that in Europe, data sovereignty isn't a preference but a procurement requirement, and that one signed pilot with a recognizable customer is worth more than any pitch deck.

What's next for Optim

Closing the first signed European pilot to convert founder networks into visible market validation. Doubling down on the autonomous routing and optimization engine as our defensible core, while resisting drift toward "just a better reporting layer." Expanding the FinTech-specific framing — model arbitrage, AI efficiency ratios, fiat-equivalent valuation — to position Optim as financial control infrastructure, not an AI tool.

Built With

• cerebras
• gcp
• next.js
• node.js