From 5ea358643181baae0d3624847e36bb0d45f0af5b Mon Sep 17 00:00:00 2001
From: DBT <nvue.dev@gmail.com>
Date: Wed, 25 Feb 2026 11:13:58 +0000
Subject: [PATCH] persist node state to shared file for gateway/cli visibility

---
 pkg/agent/loop.go    |  1 +
 pkg/nodes/manager.go | 63 +++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 63 insertions(+), 1 deletion(-)

diff --git a/pkg/agent/loop.go b/pkg/agent/loop.go
index befcf18..41812f7 100644
--- a/pkg/agent/loop.go
+++ b/pkg/agent/loop.go
@@ -88,6 +88,7 @@ func NewAgentLoop(cfg *config.Config, msgBus *bus.MessageBus, provider providers
 	toolsRegistry.Register(tools.NewProcessTool(processManager))
 	nodesManager := nodes.DefaultManager()
 	nodesManager.SetAuditPath(filepath.Join(workspace, "memory", "nodes-audit.jsonl"))
+	nodesManager.SetStatePath(filepath.Join(workspace, "memory", "nodes-state.json"))
 	nodesManager.Upsert(nodes.NodeInfo{ID: "local", Name: "local", Capabilities: nodes.Capabilities{Run: true, Invoke: true, Model: true, Camera: true, Screen: true, Location: true, Canvas: true}, Models: []string{"local-sim"}, Online: true})
 	nodesManager.RegisterHandler("local", func(req nodes.Request) nodes.Response {
 		switch req.Action {
diff --git a/pkg/nodes/manager.go b/pkg/nodes/manager.go
index 55b2b87..5d17796 100644
--- a/pkg/nodes/manager.go
+++ b/pkg/nodes/manager.go
@@ -21,6 +21,7 @@ type Manager struct {
 	handlers  map[string]Handler
 	ttl       time.Duration
 	auditPath string
+	statePath string
 }
 
 var defaultManager = NewManager()
@@ -39,6 +40,13 @@ func (m *Manager) SetAuditPath(path string) {
 	m.auditPath = strings.TrimSpace(path)
 }
 
+func (m *Manager) SetStatePath(path string) {
+	m.mu.Lock()
+	m.statePath = strings.TrimSpace(path)
+	m.mu.Unlock()
+	m.loadState()
+}
+
 func (m *Manager) Upsert(info NodeInfo) {
 	m.mu.Lock()
 	now := time.Now().UTC()
@@ -60,15 +68,21 @@ func (m *Manager) Upsert(info NodeInfo) {
 	}
 	m.nodes[info.ID] = info
 	m.mu.Unlock()
+	m.saveState()
 	m.appendAudit("upsert", info.ID, map[string]interface{}{"existed": existed, "endpoint": info.Endpoint, "version": info.Version})
 }
 
 func (m *Manager) MarkOffline(id string) {
 	m.mu.Lock()
-	defer m.mu.Unlock()
+	changed := false
 	if n, ok := m.nodes[id]; ok {
 		n.Online = false
 		m.nodes[id] = n
+		changed = true
+	}
+	m.mu.Unlock()
+	if changed {
+		m.saveState()
 	}
 }
 
@@ -208,6 +222,9 @@ func (m *Manager) reaperLoop() {
 			}
 		}
 		m.mu.Unlock()
+		if len(offlined) > 0 {
+			m.saveState()
+		}
 		for _, id := range offlined {
 			m.appendAudit("offline_ttl", id, nil)
 		}
@@ -234,3 +251,47 @@ func (m *Manager) appendAudit(event, nodeID string, data map[string]interface{})
 	defer f.Close()
 	_, _ = f.Write(append(b, '\n'))
 }
+
+func (m *Manager) saveState() {
+	m.mu.RLock()
+	path := m.statePath
+	items := make([]NodeInfo, 0, len(m.nodes))
+	for _, n := range m.nodes {
+		items = append(items, n)
+	}
+	m.mu.RUnlock()
+	if strings.TrimSpace(path) == "" {
+		return
+	}
+	_ = os.MkdirAll(filepath.Dir(path), 0755)
+	b, err := json.MarshalIndent(items, "", "  ")
+	if err != nil {
+		return
+	}
+	_ = os.WriteFile(path, b, 0644)
+}
+
+func (m *Manager) loadState() {
+	m.mu.RLock()
+	path := m.statePath
+	m.mu.RUnlock()
+	if strings.TrimSpace(path) == "" {
+		return
+	}
+	b, err := os.ReadFile(path)
+	if err != nil {
+		return
+	}
+	var items []NodeInfo
+	if err := json.Unmarshal(b, &items); err != nil {
+		return
+	}
+	m.mu.Lock()
+	for _, n := range items {
+		if strings.TrimSpace(n.ID) == "" {
+			continue
+		}
+		m.nodes[n.ID] = n
+	}
+	m.mu.Unlock()
+}