Peer-to-peer GPU mesh
Decentralized compute for AI
OpenTela pools GPUs across a peer-to-peer network into one OpenAI-compatible endpoint — with no central coordinator.
Features
Everything you need for decentralized inference
Peer-to-Peer Orchestration
Decentralized resource scheduling over a libp2p mesh network. No single point of failure.
GPU-Native
Built for GPU workloads. Auto-detects hardware via nvidia-smi and integrates with Slurm clusters.
CRDT Consensus
Conflict-free replicated state with no central coordinator. Gossip-based propagation across all peers.
Smart Routing
Three-tier request matching with automatic fallback. Route by model, capability, or catch-all.
OpenAI-Compatible API
Drop-in replacement for OpenAI endpoints. Point your existing apps at an OpenTela cluster.
News
From the OpenTela blog
How SwissAI Uses OpenTela for Scalable LLM Serving
How the Swiss AI Initiative leverages OpenTela to serve LLMs on HPC clusters, enabling researchers to access powerful models seamlessly.
Read more →ResearchReproducing the Trace Analysis Figures from our OSDI '26 Paper
Artifact guide for the trace-analysis figures in our OSDI '26 paper — find the SwissAI serving trace and regenerate the plots from the OpenTela repository.
Read more →Community
Get involved