From ORM plumbing to LLM plumbing
Luca Steeb's GitHub profile tells the story before any interview does. Previous experience at prisma.io — the company that built the next-generation ORM for Node.js — and at hatchet, the orchestration engine for background tasks and AI agents. Then consulting for IT companies on clean code and scalable tech stacks. A career spent in infrastructure: the unglamorous layer that everything else depends on.
That background matters because LLM Gateway is infrastructure in the purest sense. As AI models multiplied across 2024 and 2025 — OpenAI, Anthropic, Google, and dozens more — every development team faced the same mess: different APIs, different SDKs, different pricing, different failure modes. The problem rhymed exactly with what Prisma had solved for databases: one interface over many backends.
Steeb teamed up with Ismail, known online as smakosh, and the two started building. In a DevNTell podcast episode (EP 180), they walked through the origin: developers needed middleware between their applications and the exploding universe of LLM providers, and nothing open-source quite did the job.
The founder-market fit is almost too clean. Steeb had spent years watching developers struggle with provider sprawl at Prisma and Hatchet. He wasn't guessing at the pain — he'd been in the rooms where teams debated which SDK to standardize on.
One endpoint to rule the models
LLM Gateway's pitch fits in a sentence: a unified, OpenAI-compatible API endpoint that routes requests to any large language model provider. Change the base URL, keep your existing code — whether it's cURL, Python, TypeScript, or Go — and switch models across providers by changing one string.
Under the hood, it's a TypeScript monorepo with real operational depth: smart retries with automatic fallback when a provider fails, centralized API key management so provider credentials stay server-side, real-time cost and token tracking per request, performance monitoring to compare models on accuracy versus cost, and team management with guardrails.
The open-source decision was structural, not just philosophical. The core is AGPL-3.0 licensed, which means developers can self-host on their own infrastructure — a critical requirement for teams with data residency or compliance constraints. Enterprise features sit behind a commercial license, creating the classic open-core monetization ladder: free self-hosted core, paid hosted service and advanced features.
This is the same playbook that built Prisma, incidentally. Open-source the core, make it genuinely useful self-hosted, then monetize the teams that want someone else to run it. Steeb had a front-row seat to that model working — then applied it to the hottest infrastructure category of the decade.
Building in public, shipping in the open
The DevNTell appearance is representative of how LLM Gateway grows: developer-to-developer, in public. The episode walked through a live demonstration — self-hosting, integration, the roadmap — with timestamps for each section. No polished marketing, just two founders showing the thing working.
The GitHub repository backs up the talk. Active development, clear AI gateway positioning, Docker-based self-hosting, dashboard apps for usage analytics — it's a real project with real users, not a landing page with a waitlist. The awesome-llm-gateways community list ranks it among the notable open-source gateways, citing its centralized key management, token and cost tracking, and operational visibility as differentiators over thinner request proxies.
Steeb's consulting background shows in the product's pragmatism. Features like per-key monthly spend caps (returning 402 when exceeded), role-based access controls, and automatic fallback chains aren't exciting — they're what production teams actually need before they'll trust a gateway with real traffic. Infrastructure buyers don't want vision; they want the thing to not break at 3 AM.
The 200+ models across 20+ providers figure is the compounding moat. Every new provider integration makes the gateway more valuable to existing users and harder to leave — switching costs grow with each model your team routes through it.
The playbook
- Solve the problem you watched for years — Steeb's time at Prisma and Hatchet meant he'd seen provider sprawl up close. Infrastructure founders win by recognizing rhyming problems in new categories.
- Open-source the core, monetize the operations — AGPL-3.0 self-hosting earns developer trust; hosted service and enterprise features earn revenue. The Prisma playbook, applied to LLMs.
- Build in public with developers — Podcast demos, active GitHub, community lists. Developer tools grow through credibility, not ad spend.
- Ship boring reliability features — Spend caps, retries, fallback chains, key management. Production teams buy the unglamorous stuff that prevents 3 AM pages.
- Compound with integrations — Every new provider makes the gateway stickier. Switching costs grow automatically as teams route more models through one endpoint.