Router that virtually distributes inference across connected devices in the home.
Go
Correct worker counts after the engine proxy unification
Terve12bf25644c3c
diff --git a/AGENTS.md b/AGENTS.mdindex 28c4813..db8908c 100644--- a/AGENTS.md+++ b/AGENTS.md@@ -43,7 +43,7 @@ change belongs in a Go service instead.- Electron starts **only** `nvpair-ui-broker` from `desktop/cli-bin/`.- The broker supervises the other Go workers: discovery, proxies, engines,cluster, settings, manual nodes, workloads, errors, and the scheduler.-- The broker spawns all 11 workers at startup. Only the scanner is required; the+- The broker spawns all 10 workers at startup. Only the scanner is required; therest are optional and non-fatal.- `nvpair-tui` is bundled but never supervised. It owns its own broker.- Interprocess communication is newline-delimited JSON-RPC 2.0 over stdiodiff --git a/services/nvpair-cluster-manager/README.md b/services/nvpair-cluster-manager/README.mdindex 80d2e00..10ca938 100644--- a/services/nvpair-cluster-manager/README.md+++ b/services/nvpair-cluster-manager/README.md@@ -8,7 +8,7 @@ SPDX-License-Identifier: Apache-2.0> This README is the **integrator's guide**: how a parent process drives> cluster membership and node pairing over this component's JSON-RPC> protocol. The binary is built by `build.sh` / `build.bat` alongside the-> other twelve components and is spawned by `nvpair-ui-broker` at startup,+> other eleven components and is spawned by `nvpair-ui-broker` at startup,> as an optional worker — a missing binary leaves the broker running without> cluster pairing.diff --git a/services/nvpair-ui-broker/README.md b/services/nvpair-ui-broker/README.mdindex df5ac9d..91cd7f9 100644--- a/services/nvpair-ui-broker/README.md+++ b/services/nvpair-ui-broker/README.md@@ -15,7 +15,7 @@ the broker from the same installation directory. The broker is the parentprocess and canonical backend entry point: it supervises the worker subprocesseson the UI's behalf, speaking JSON-RPC over stdio.-The broker supervises **eleven** worker subprocesses, so a client gets the whole+The broker supervises **ten** worker subprocesses, so a client gets the wholebackend behind one endpoint. Each is spawned at startup and relayed under its ownnamespace:@@ -23,8 +23,7 @@ namespace:| --- | --- | --- || `nvpair-node-scanner` | Discovery daemon: advertises this host's one `_nvpair-node._tcp` record and browses the LAN | `discovery:*` || `nvpair-node-info` | Local GPU / CPU / memory inventory over HTTP at `/v1/node-info` | — (HTTP only) |-| `ollama-proxy` | Ollama-compatible inference proxy and router | `ollama-proxy:*` |-| `lmstudio-proxy` | The LM Studio counterpart, supervised identically | `lmstudio-proxy:*` |+| `nvpair-proxy` | One process hosting an inference proxy and router facade per enabled engine | `ollama-proxy:*`, `lmstudio-proxy:*` || `nvpair-engine-manager` | Local engine and model control plane; also serves `GET /v1/models` to peers | `engine:*` || `nvpair-cluster-manager` | Node identity, trusted-node store, PIN pairing | `cluster:*`, `nodes:*` || `nvpair-workload-manager` | Cluster workload relay between this node and peers | `workloads:*` |