{"slug": "buzz-share-spare-compute-and-run-open-models-together-over-p2p-networks", "title": "Buzz: Share spare compute and run open models together over P2P networks", "summary": "Block is building Buzz, an open-source tool that lets community members share spare compute and run open models together over peer-to-peer networks using iroh's networking stack. Buzz integrates MeshLLM, which connects members' machines via iroh so local agents can use shared inference capacity through an OpenAI-compatible endpoint, with sharing entirely opt-in and the inference path bypassing the Buzz relay. Buzz Mesh gates shared inference on existing community membership, requiring a valid endpoint binding, an invite token that decodes to a bounded iroh endpoint, and advertised relay URLs matching the machine's local iroh relay policy before a serving target becomes selectable.", "body_md": "# Buzz: share spare compute and run open models together over p2p networks.\n\nby Rae McKelvey\nHow can we use the devices we already have to build bigger things with AI?\nAs hardware gets more expensive, that question is becoming an important new\nfrontier for software engineers. Block is building\n[Buzz](https://github.com/block/buzz) to chip away at the problem, powered by\niroh's peer-to-peer networking.\n\nAgents are starting to do real work: opening repositories, writing patches, reviewing code, running workflows, joining conversations, and carrying context across a team. Once that happens, a workspace is no longer just a place where humans talk. It is a network of computers whose resources can become part of the development infrastructure.\n\nThat is what Block is building with Buzz. People can bring their own agents or use the Buzz agent, but either way those agents need an inference provider. A Buzz community creates a natural perimeter for members who want to share spare compute and run open models together.\n\n[MeshLLM](https://www.iroh.computer/blog/mesh-llm) is built into Buzz. It uses iroh to\nconnect community members across all sorts of networks and make their compute\navailable to local agents without putting a central server in the inference\npath. Sharing is entirely opt-in. When enabled, it can reduce inference costs\nand put hardware that would otherwise sit idle to useful work.\n\nA member can opt in to share compute, and their machine advertises the models it can serve. Other members' agents use that capacity through an OpenAI-compatible local endpoint. The Buzz relay helps the community discover who is allowed to participate, but the inference path does not run through the Buzz relay.\n\n## [Buzz Mesh: your community is your compute](#buzz-mesh-your-community-is-your-compute)\n\nIn Buzz, a community is already a trust group: the people and agents admitted to a network. Buzz Mesh reuses that decision. The same membership that lets someone read channels and participate in work also gates shared inference.\n\nCurrent Buzz membership controls which node identities are admitted. A serving target becomes selectable only when its endpoint binding is valid, its invite token decodes to a bounded iroh endpoint, and every advertised relay URL matches the machine's locally configured iroh relay policy. The relay is not blindly handing clients arbitrary network targets. Discovery is member-signed, endpoint-bound, size-bounded, and policy-checked before a node becomes selectable.\n\nThe result is a clean split:\n\n| Piece | Responsibility | \n|---|---|\n| Buzz relay | Membership, signed status, discovery, and workspace history | \n| MeshLLM | Inference transport, owner allowlists, and model routing | \n| iroh | Authenticated QUIC connections and direct network paths | \n\n## [Why iroh](#why-iroh)\n\nBuzz does not want to require a VPN, a new global network account, or every team to become network operators before two machines can talk.\n\n[Libp2p could also work, but](https://www.iroh.computer/blog/comparing-iroh-and-libp2p) Buzz needs a smaller primitive that handles the network and gets out of the way:\ndial an identity, open authenticated streams, and let the application own the\nprotocol above them.\n\nEvery iroh endpoint is already identified by a public key. The address is not\n\"this IP and port\"; it is \"this endpoint.\" That is exactly the shape an agent\nworkspace wants. Buzz decides who may enter the community. MeshLLM decides who\nmay serve or consume inference. iroh makes the connection work across laptops,\nworkstations, home and office networks, and cloud machines through\n[global endpoint address lookup](https://docs.iroh.computer/concepts/address-lookup).\n\nThe network becomes something MeshLLM and Buzz can build in directly, without asking the user to configure it.\n\n## [How Buzz uses iroh](#how-buzz-uses-iroh)\n\nThe desktop runtime publishes a signed, replaceable discovery note containing the member's MeshLLM owner identity and current iroh endpoint. Other members resolve available serving targets from the relay, validate the endpoint and relay policy locally, and then connect through MeshLLM.\n\nTo an agent, it still looks like an OpenAI-compatible endpoint at\n`127.0.0.1:9337/v1`. The model can be a specific name or `auto`, which Buzz maps\nto MeshLLM's virtual `mesh` model so the mesh can choose the best available\nroute.\n\nBuzz also has an inter-relay mesh crate built around one iroh endpoint per relay runtime. Each runtime gets a unique iroh keypair. Peers exchange attested readiness records, gossip membership, and move fenced tunnel traffic over QUIC streams and datagrams.\n\nThe wire contract is explicit:\n\n| Piece | Purpose | \n|---|---|\n| `buzz/mesh/1` | ALPN for Buzz relay mesh connections | \n| QUIC bidirectional streams | Reliable tunnel traffic and control | \n| QUIC datagrams | Real-time media, including huddle audio | \n| `FencedHeader` | `{session_id, generation, owner_runtime_id}` on session-bearing frames | \n| Redis CAS lease | Arbiter for ownership; mesh membership is only a routing hint | \n\nThat distinction matters. The mesh can say, \"this peer is dialable.\" It cannot say, \"this peer owns the session.\" Ownership is fenced by the relay's session directory, and stale generations are rejected at every hop.\n\nThat is the kind of boring engineering detail that makes distributed systems usable.\n\n## [What's next](#whats-next)\n\n[Goose](https://github.com/block/goose) now has an optional feature called\n`roam`, which allows agents to exchange cards and talk directly. The next step\nis more agent-to-agent communication over iroh with little or no user\nconfiguration.\n\nBuzz provides a natural grouping of people, their agents, and their compute. Much of that compute may be idle. MeshLLM lets the group share it over iroh.\n\nThe larger question is profound: how can we use the devices we already have to build bigger things with AI? As hardware costs rise, Buzz and MeshLLM are working on one answer—wholeheartedly, and powered by peer-to-peer networks.\n\nTo get started, take a look at our\n\n[docs](https://iroh.computer/docs), dive directly into\n\n[the code](https://github.com/n0-computer/iroh), or chat with us in our\n\n[discord channel](https://iroh.computer/discord).", "url": "https://wpnews.pro/news/buzz-share-spare-compute-and-run-open-models-together-over-p2p-networks", "canonical_source": "https://www.iroh.computer/blog/buzz-agent-workspaces", "published_at": "2026-09-23 20:03:32+00:00", "updated_at": "2026-09-23 20:31:47.984543+00:00", "lang": "en", "topics": ["ai-infrastructure", "ai-agents", "ai-tools", "artificial-intelligence"], "entities": ["Block", "Buzz", "iroh", "MeshLLM", "Rae McKelvey", "Libp2p"], "alternates": {"html": "https://wpnews.pro/news/buzz-share-spare-compute-and-run-open-models-together-over-p2p-networks", "markdown": "https://wpnews.pro/news/buzz-share-spare-compute-and-run-open-models-together-over-p2p-networks.md", "text": "https://wpnews.pro/news/buzz-share-spare-compute-and-run-open-models-together-over-p2p-networks.txt", "jsonld": "https://wpnews.pro/news/buzz-share-spare-compute-and-run-open-models-together-over-p2p-networks.jsonld"}}