diff --git a/desktop/src-tauri/src/commands/agents.rs b/desktop/src-tauri/src/commands/agents.rs
index c666af6632..4d667e83e3 100644
--- a/desktop/src-tauri/src/commands/agents.rs
+++ b/desktop/src-tauri/src/commands/agents.rs
@@ -44,6 +44,17 @@ fn build_deploy_payload(
         crate::managed_agents::resolve_persona_env(app, record.persona_id.as_deref())?;
     let merged_env = crate::managed_agents::merged_user_env(&persona_env, &record.env_vars);
 
+    // Resolve system prompt and model via the persona (same as local spawn).
+    // Re-reads the persona store so edits take effect without recreating the agent.
+    let personas = load_personas(app).unwrap_or_default();
+    let (effective_prompt, effective_model) =
+        crate::managed_agents::resolve_effective_prompt_and_model(
+            record.persona_id.as_deref(),
+            &personas,
+            record.system_prompt.as_deref(),
+            record.model.as_deref(),
+        );
+
     Ok(serde_json::json!({
         "name": &record.name,
         "relay_url": &record.relay_url,
@@ -51,8 +62,8 @@ fn build_deploy_payload(
         "auth_tag": &record.auth_tag,
         "agent_command": &record.agent_command,
         "agent_args": &record.agent_args,
-        "system_prompt": &record.system_prompt,
-        "model": &record.model,
+        "system_prompt": &effective_prompt,
+        "model": &effective_model,
         "turn_timeout_seconds": record.turn_timeout_seconds,
         "idle_timeout_seconds": record.idle_timeout_seconds,
         "max_turn_duration_seconds": record.max_turn_duration_seconds,
diff --git a/desktop/src-tauri/src/managed_agents/runtime/tests.rs b/desktop/src-tauri/src/managed_agents/runtime/tests.rs
index 446a6ba28d..d018380490 100644
--- a/desktop/src-tauri/src/managed_agents/runtime/tests.rs
+++ b/desktop/src-tauri/src/managed_agents/runtime/tests.rs
@@ -216,8 +216,18 @@ fn resolve_prompt_falls_back_when_no_persona_id() {
     assert_eq!(model.as_deref(), Some("record-model"));
 }
 
+/// Tests prompt/model resolution **in isolation** when the persona is missing.
+///
+/// In a real spawn, `resolve_persona_env()` (called later in `spawn_agent_child`)
+/// would **fail closed** if the persona cannot be found — because persona env_vars
+/// may contain required API credentials. This test only validates the
+/// `resolve_effective_prompt_and_model` helper's fallback behavior, NOT that a
+/// full agent spawn would succeed with a missing persona.
 #[test]
 fn resolve_prompt_falls_back_when_persona_not_found() {
+    // NOTE: This tests the prompt/model resolution layer only. A full spawn
+    // with a missing persona_id would fail at resolve_persona_env() before
+    // the process is ever created.
     let personas = vec![persona_fixture("custom:other", "Other prompt", None)];
 
     let (prompt, model) = resolve_effective_prompt_and_model(