Skip to content
Docs menu

Mesh

How Venom Cloud sends requests to GPUs you own (your key groups' daemons) or to public nodes run by anyone, while provider keys and the data path stay on those machines.

View .md

The mesh is how Venom Cloud reaches machines it does not run. A daemon is venom's Go orchestrator with mesh mode on: a home GPU, a VM, a laptop or a rented GPU box. It dials out to the cloud, takes the requests the cloud routes to it, and serves them with its own upstreams (vLLM, SGLang, llama.cpp, Ollama, provider keys). Clients see one OpenAI-compatible /v1 at the cloud.

 clients (venom CLI, any OpenAI SDK)            Venom Cloud                      daemons (dial out only)
 VENOM_BASE_URL=https://<cloud>/v1  ──POST──▶  auth, limits, budget, router ──▶ go-orchestrator + mesh mode
                                    ◀─307 + ticket (direct) ─┘   │ relay            ├─ local /v1 (unchanged)
                                    ◀═══ tokens via the cloud ═══╛                  └─ engines and provider keys
                                    ═══ tokens straight from the daemon (direct) ══▶

Three rules hold everywhere:

  • Daemons only dial out. No inbound port or public IP is needed. A daemon that also publishes an endpoint the client can reach gets the faster direct path.
  • Keys stay local. Provider keys and engine URLs never leave the daemon's machine. The cloud sees the model list a daemon advertises, its load, and the usage it reports.
  • Mesh mode is opt-in. An orchestrator does nothing with the cloud until you set VENOM_CLOUD_URL and join with a token, or connect from the desktop daemon's dashboard. Its local /v1 keeps serving either way.

Private mesh and public mesh

Private mesh Public mesh
Who runs the daemons your org, one key group per daemon (or key groups that share with you) anyone with a public-pool join token
Who may use them keys of the key group, or key groups it shares with keys with min_trust_class: public, or an org routing policy that allows public nodes
Path direct (307 + ticket) or relayed relayed only, so the cloud witnesses every job
Trust the operator is your own org signed receipts, sampled audits, canaries, witnessed QoS
Engines anything the orchestrator fronts vLLM or SGLang with venom-sampler/1, llama.cpp through the venom-llamacpp shim; no Ollama

Pages

To Read
Understand routing, tickets, the relay and what the cloud sees How the private mesh works
Put a machine on the mesh Connect a daemon
Host a public node, or use public nodes The public mesh
Check what a public node served, or who made a commit Provenance and receipts
Write or debug a daemon against the wire protocol Daemon API reference

In this section

  • How the private mesh works: How Venom Cloud routes a key group's requests to its own daemons, directly with a 307 and a ticket or relayed through the cloud, and what each side sees, signs and keeps.
  • Connect a daemon: Put a machine on your org's Venom Cloud mesh with a join token, from the desktop daemon's dashboard, make daemon, or the orchestrator binary, and set its endpoints, location and limits.
  • The public mesh: Host a public node that serves strangers' requests, or route your requests to public nodes, and how signed receipts, sampled audits, canaries and witnessed QoS keep nodes honest.
  • Provenance and receipts: The two kinds of signed receipt venom produces, for a completion a public node served and for a commit an agent made, what each proves, and how to check them yourself.
  • Daemon API reference: Every route a daemon calls on Venom Cloud under /api/mesh/v1, with its authentication, request and response bodies, and errors.