An engineering practice, not a platform pitch
2oo.one exists because the organisations with the most to gain from AI — hospitals, law firms, banks, agencies, manufacturers — are the ones least able to send their data to someone else's servers.
What we believe
Open-weight models have closed most of the quality gap with cloud APIs for enterprise work. What organisations lack is not models — it is engineering: sizing, quantization, serving, evaluation, integration and disciplined operations. That is the work we do.
We are deliberately not a platform vendor. There is no license to renew, no per-seat fee and no telemetry phoning home. Every engagement ends with you owning the stack and the knowledge to run it.
How we stay honest
- We tell you when APIs are cheaper for your volume — and show the maths.
- Every model decision is backed by evaluation on your tasks, not leaderboards.
- Deliverables are documented for handover from day one.
- We publish our reasoning openly in our guides and calculator.
Who is behind 2oo.one
2oo.one is a venture of ESS ENN Associates and part of the No Ifs Only Bots group. Website design by 530 Expert.
What we work with
Models
Llama, Mistral, Qwen, DeepSeek, Gemma and domain-specific open models.
Serving
vLLM, TensorRT-LLM, Ollama, llama.cpp; Kubernetes or bare metal.
Customisation
LoRA, QLoRA, SFT, DPO; retrieval pipelines and evaluation harnesses.
Retrieval
Vector databases (Qdrant, Milvus, pgvector), embeddings, rerankers and GraphRAG knowledge-graph pipelines.
Governance
SSO (SAML/OIDC), usage metering, quota and budget management, audit logging, PII guardrails.
Hardware
NVIDIA data-centre and workstation GPUs, sized to workload — vendor-neutral advice.