Model-agnostic LLM serving infrastructure — the Kubernetes-native control plane that lets enterprises run open-weight language models in production without locking into a single cloud, model vendor, or inference runtime.
The category did not exist as a named sub-segment of AI infrastructure five years ago. Today it is one of the most strategically contested layers of the enterprise AI stack, and the UK does not yet have a domestic leader in it.
Production LLM deployment in 2026 faces three independent forms of lock-in:
- Cloud lock-in — your inference fleet lives on one cloud's GPUs (AWS, GCP, Azure); moving a model means re-implementing the serving stack
- Model-vendor lock-in — your control plane is bound to one MLOps platform; moving means losing audit, lineage, and fair-share scheduling
- Runtime lock-in — your model is wired to one inference runtime (vLLM, TGI, Triton, SGLang); moving means re-tuning latency and throughput from scratch
Each lock-in is independent. Each has a different cost to break. None of the existing platforms solves all three. PYXIS3 is the operations layer that lets you mix and match — the same control plane drives the same models on the same KPIs across heterogeneous fleets: cloud, on-prem, mixed.
| Repository | Purpose | License |
|---|---|---|
pyxis-arch |
Architecture thesis — Funnel-Engine-Lens model, runtime-adapter abstraction, fair-share GPU scheduling, audit/lineage across runtime migration | MIT |
vllm-bench |
TTFT + TPOT benchmarking for OpenAI-compatible inference endpoints — vLLM, TGI, llama.cpp, Ollama | MIT |
llm-serving-cookbook |
Production recipes for Kubernetes-native vLLM serving — vLLM-on-EKS, KEDA autoscaling, token economics, TTFT optimisation, runtime selection | Apache-2.0 |
lens |
In-cluster Kubernetes observability with LLM-endpoint discovery for vLLM / TGI / llama.cpp / Ollama. In-browser kubectl exec. Live at lens.pyxis3.ai |
MIT |
noor |
Multilingual AI semantic search over Qur'an + classical Arabic literature. Sentence-Transformers + sqlite-vec. Live at noor.pyxis3.ai |
MIT |
awesome-model-agnostic-llm |
Curated index of model-agnostic LLM tooling — serving runtimes, routers, evaluators, standards | CC0-1.0 |
- Website: pyxis3.ai
- Email: ops@pyxis3.ai
- LinkedIn: /company/pyxis3
- Founder: Omar Abdrabo (@oabdrabo)
