InferLayer is an applied AI engineering firm. We make AI infrastructure spend something you can verify, not estimate. We build agent workflows, retrieval systems, and production AI products for clients, and we operate our own inference gateway to prove the case ourselves.
The gateway is live. It caches repeat requests, routes each one to the model that actually fits, and records every result in a tamper-evident, hash-chained audit log. We're taking on design partners selectively before opening it up more broadly.
Most of what we do is custom engineering work for clients. The gateway is the one piece we productized ourselves, because our own client work needed it.
The gateway sits between your application and whichever model provider you use. Every request passes through cache, classification, and cryptographic logging before it reaches a model. Integration is a one-line change to your base URL.
Most AI stacks can tell you what they spent. Few can prove it to someone outside the team. Every call through InferLayer produces a record your auditors can check on their own, without access to our systems and without taking our word for anything.
Built to support the audit-control requirement in HIPAA, SOC 2 evidence collection, GDPR accountability, and the EU AI Act's logging and log-retention duties for high-risk systems.
Token costs are one problem. The bigger one is that nobody knows where the money went. Enterprises run AI workflows across multiple providers with no real way to tie spend to a workflow, a business unit, or an outcome. The CFO sees a bill. The engineering team sees usage logs. Nobody sees both at once.
InferLayer sits between your applications and your inference providers as the layer that actually attributes spend. It routes each request to the right model, caches repeat queries, and hash-chains every log row, so the cost numbers hold up without anyone having to take our word for it.
The gateway is live in production, the audit chain is verified and running, and the benchmarks are real: roughly 50% measured cost reduction against a frontier-model baseline, with lower latency, on the default CPU tier. We're onboarding design partners now. What you're running shapes what we build next.
No open roles listed right now. If you're an inference engineer, an AI infrastructure researcher, or a systems developer who cares about getting cost attribution right, send us what you've built. We read everything.
base_url at https://api.inferlayer.ai/v1 and use an il- API key. Streaming, function calling, and tool use work without modification, same SDKs, same response shape.
base_url in any OpenAI-compatible client, confirm the first request shows up in your dashboard, done. BYOC deployment: typically 5 to 14 days, depending on your cloud environment and security review. Covers Docker Compose setup, audit log verification, dashboard configuration, and workflow-attribution tagging.
We're talking to design partners: teams running production LLM workloads who care about proving cost, not just cutting it. If that's you, send us a sample of your logs and we'll run a real diagnostic.
We reply personally. No automated sequences.