4ROUTER / AI INFRASTRUCTURE
From multi-model inference to training, GPU clusters and private deployment. Bring AI into production with the models and infrastructure your work needs.
THE PLATFORM
Keep the way you build. Bring model access, routing decisions and usage management into one platform.
from openai import OpenAI
import os
client = OpenAI(
base_url=os.environ["FOURROUTER_BASE_URL"],
api_key=os.environ["FOURROUTER_API_KEY"]
)
models = client.models.list()OPENAI-COMPATIBLE API
Change one base_url to integrate; behind the endpoint is the full model pool.
ILLUSTRATIVE REQUEST PATH
ENTHALPY ROUTING ENGINE
Enthalpy, our routing engine, brings task requirements, model capabilities and cost constraints into each decision.
Engine details →Request-level usage records help you trace calls and reconcile costs.
Manage accounts and access through the console, with permissions scoped to each workspace.
CAPABILITIES / SERVICES
Build on the platform, or work with us on infrastructure of your own.
INFERENCE
Access multiple models through one endpoint, or deploy model serving in a dedicated environment. Match models, capacity and access to the workload.
Inference in detail →Connect providers through a compatible interface, with less integration work in your application.
Configure dedicated serving around your model, concurrency and data requirements.
TRAINING & OPTIMISATION
Fine-tuning, evaluation and kernel optimisation for a defined task. Start with quality and operational requirements, then choose the tuning path.
Training and tuning in detail →Fine-tune around your data and task, and evaluate the result on a consistent basis.
Find bottlenecks from model execution down to kernel implementation, then optimise what matters.
YOUR ENVIRONMENT
When you need your own compute or defined data boundaries, our work extends to the cluster and the deployment.
From hardware selection to scheduling and monitoring, the deliverable is a cluster that can be operated.
Deploy models and serving in your chosen environment, with explicit boundaries for data, access, updates and operations.
4ROUTER / SYSTEMS RESEARCH
Understand how operators, KV state and scheduling shape an inference path.
Read the systems notes in Resources ↗Identify the limiting resource before selecting the implementation.
↗ATTENTION / MEMORYOne optimizes data movement; the other organizes persistent state.
↗TIERS / TRANSFERAvoiding prefill requires paying for state access.
↗4ROUTER / ENGINEERING
Systems
From individual GPUs to a coordinated serving system.
Performance
Inside the work of balancing latency and throughput.
Serverless
How workloads become services that scale on demand.
Routing
Tracing a request through selection and dispatch.
Systems
Placing work where the right resources are available.
Performance
The state behind generation, and how it uses memory.
Performance
Coordinating generation as requests arrive and finish.
Serverless
Understanding the path from launch to readiness.
Serverless
Matching serving capacity to the shape of demand.
Routing
Defining alternate paths and the limits of recovery.
Observability
Connecting admission, execution and accounting.
Reliability
Separating resources, data and failure domains.
Connect to the platform, or discuss models, performance and deployment.