Lumabri
Run Moe Models on a P2P Swarm with Colibri
Get picks like this daily. Today's launch numbers, the top launches, AI and tech news, and a daily opportunity pick, straight to your inbox.
Want to reach these readers? Sponsor an issue.
HN · launched Aug 14, 2026 · 48 upvotes (top 14%) · 0 views
Niche: machine learning inference engines · 108 launches, 8 in the last 7 days ▼ -27% · Explore this niche →
Category and tags
- Horizontal
- Model & infra
- Developer
- AI-native
- Hobby / open-source project
Visit website ↗ Source listing ↗
More like this: all 108 launches in machine learning inference engines →
Could you build this?
No Running decentralized Mixture-of-Experts (MoE) LLM inference across a peer-to-peer swarm demands specialized distributed systems engineering, tensor parallelization, and P2P networking protocols.
What it would actually take: Architecture requires a custom peer-to-peer transport protocol (e.g., libp2p) combined with distributed dynamic tensor parallelism (similar to Petals or Hivemind). Hard challenges include latency-sensitive pipelined expert routing across asymmetric consumer internet connections, fault tolerance against dynamic peer dropouts, and custom CUDA/Triton kernels for decentralized MoE execution.
Discussion
19 comments analyzed.
Competitors mentioned: MeshLLM, llama.cpp
Concerns raised: Floating point determinism issues across different hardware and instruction sets, High latency for inference, especially over internet vs local networks, Accumulation order differs across CPU/GPU peers in same swarm, Lack of concrete performance benchmarks and statistics in documentation
Feature requests: Add performance stats and experiment logs (throughput, latency, bandwidth per model), Integrate HuggingFace model support directly like llama.cpp does, Use specific model repository names instead of vague model identifiers, Support for arbitrary layer chunking to increase memory throughput, Support for IoT/busybox devices contributing to inference swarms
Competitors
Other products that read as similar to this one : 2537 launches clear the similarity bar, closest 8 shown.
Attention rank: #302 of 2538 (itself plus its competitors, highest first, normalized so YC and Product Hunt are compared fairly).
Launched 289 days after the earliest competitor.
- OpenRaid · github · 2026-10-03 · 8 upvotes · similarity 0.64
- swarmchat · github · 2026-10-29 · 46 upvotes · similarity 0.60
- A tool to manage a swarm of coding agents on Linux · hn · 2026-04-10 · 5 upvotes · similarity 0.59
- Swarmmy · github · 2026-09-22 · 22 upvotes · similarity 0.58
- Acyclic Labs: Building the foundations for AI swarms · yc · 2026-11-13 · 23 upvotes · similarity 0.57
- scp · github · 2026-10-18 · 65 upvotes · similarity 0.57
- SwarmClaw · hn · 2026-03-01 · 5 upvotes · similarity 0.56
- fleet · github · 2026-10-09 · 97 upvotes · similarity 0.55
The niche, labels, vibe code rating and competitor comparison on this page are analysis by Nicheloom. If you use them, please credit nicheloom.com. How you may use our data.