OhhO Serve
Robot AI inference, as an API.
Robot AI inference, as an API. Deploy a Vision-Language-Action model behind a REST endpoint in one command — pluggable model backends, batching, and Prometheus metrics built in.
- One-command model deploy
- REST predict API
- Pluggable backends (OpenVLA, SmolVLA, ACT…)
- 4-bit quantization option
- Prometheus metrics built in
What you get
Modern robots run on large Vision-Language-Action (VLA) models, but getting one into production means GPU servers, model loading, batching, versioning and monitoring. OhhO Serve turns all of that into a single endpoint.
Point Serve at a model — OpenVLA, SmolVLA, ACT, a diffusion policy, or your own fine-tune — and it exposes a clean REST API your robot calls with an image and an instruction, and gets back an action. Swap models without touching robot code.
Serve is built for the real world: health checks, hot model loading, 4-bit quantization for tighter VRAM budgets, and first-class Prometheus metrics so you can see latency and throughput at a glance.
Built to do the hard parts for you
OpenVLA, SmolVLA, ACT, diffusion, or a custom class — selected by config, not code.
POST an image + instruction, get back an action vector. Your robot doesn't need to know which model is behind it.
/health, /load_model and /predict endpoints let you swap or reload models with zero downtime.
Optional 4-bit quantization runs large VLA models on modest GPUs. Bring your own — we help you size it, or recommend hardware that fits your budget.
Latency, throughput and GPU metrics export to Prometheus and the OhhO Fleet dashboards.
How it works
Set the model class and checkpoint via environment variables.
One command starts the FastAPI inference server.
Your robot sends camera + instruction and receives an action.
Metrics stream to Prometheus / Grafana; raise call limits as you grow.
Specifications
- API
- REST — /health, /load_model, /predict
- Backends
- OpenVLA, SmolVLA, ACT, diffusion, custom
- Quantization
- Optional 4-bit
- Hardware
- Your GPU (desktop, server or cloud) — we help you size it
- Metrics
- Prometheus /metrics endpoint
- Deploy
- Docker, single command
Which plan do I need for OhhO Serve?
We recommend the Fleet plan. Builder's 500 calls/day is for prototyping. If real robots are calling the model in production, choose Fleet for 10K/day — or Forge for an on-prem license with no metering.
Common questions
Yes. Serve loads any compatible checkpoint, including models you've fine-tuned with OhhO Data and the OmniVLA engine.
On your GPU — desktop, server or cloud. Serve is software; you keep the model and the data.
Works better together
Ready to build with OhhO Serve?
Start free and simulate first — no hardware required. Upgrade when you're ready to deploy.