Peer-to-peer AI infrastructure and agent software
Hyperspace
Hyperspace develops a peer-to-peer AI network that lets participants run language models and contribute compute, storage or bandwidth. Its CLI, desktop tray app and browser client provide different ways to participate: browsers use WebGPU and WebLLM, while installed clients support native GPU inference and Ollama integration. Developers and node operators can generate text, create embeddings for semantic search and provide distributed storage.
Co-founder Varun Mathur outlined an early collaborative AI approach in 2023 through Collama, a retrieval-augmented generation tool built around community-curated sources. Hyperspace’s network uses libp2p, distributed hash tables and GossipSub for peer discovery and communication. Mathur’s 2026 Proof of Intelligence paper extends the collaboration model to computational experiments: its proposed ResearchDAG records experiment configurations, metrics, execution proofs and relationships to earlier work, with rewards tied to other nodes adopting results.
Hyperspace Wallet adds payment infrastructure for agent developers through TypeScript and Python SDKs. It combines portable agent identity, payment routing, spending limits and common receipt formats; the documented end-to-end implementation runs on the A1 testnet. The SDK is free, with a stated business model based on transaction and marketplace fees. This extends Hyperspace’s scope from sharing computing resources to enabling agents to purchase inference, data and tool calls.
1 talk
Newest first1 speaker at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.
Messages from the stage
Different models for different workflow steps
The demonstration combined Qwen2 Instruct for reasoning with Llama 3 70B for summarization and executable Python outputs. Puppeteer and Beautiful Soup prepared web content for language models.
Affiliations reflect each recorded session, not necessarily current employment.
