SRF-01Near-Perfect Uptime
No centralized compute provider sits in the critical path.
Traffic shifts between healthy nodes as capacity changes.
About
We stopped routing every request through centralized compute.
We started routing inference across an independent field of compute nodes.
One endpoint enters.
The network takes it from there.
Surf observes capacity, latency and model health across an evolving field of independent compute nodes.Each request is routed to the strongest available path, then streamed back through one OpenAI-compatible interface.
- Distributed compute
- Adaptive routing
- Continuous health checks
- Open-source models
- OpenAI-compatible
Distributed Systems
SRF-01Near-Perfect Uptime
No centralized compute provider sits in the critical path.
Traffic shifts between healthy nodes as capacity changes.
SRF-02Lower Cost than Hyperscalers
Surf pools capacity across independent compute nodes, avoiding centralized infrastructure premiums.
Lower infrastructure overhead is passed through to inference pricing.
SRF-03Higher Throughput
Requests fan out across available compute nodes instead of one centralized queue.
Capacity expands with the node field as demand rises.
Open-source models, available through a shared inference network.
New models join without changing how requests enter the network.
Kimi K2.7 Code
Code-specialized model for agentic software work.
262,144 token context · reasoning · tools · code · structured-output.
Kimi K2.6
High-reasoning general model with tools and vision.
262,144 token context · reasoning · tools · vision · structured-output.
GLM 5.2
Fast general intelligence with strong tool use.
128,000 token context · reasoning · tools · vision · structured-output.
Open models should not depend on centralized compute.
Not one region.
Not one provider.
Distributed inference, observed and routed in real time.