fix(dash): discover served model for configured chat endpoint

A configured chat URL (OPENAI_BASE_URL / --chat-url) with no explicit
model resolved to the "local-model" placeholder, which 404s on servers
that register the model under its real id. Query /v1/models on startup
to adopt the served model, mirroring the managed-service path. Gated on
a reachable probe so an unreachable endpoint adds no fetch timeout to
launch, and an explicit --chat-model/config value still wins.

Signed-off-by: Eugene Volen <Eugene.Volen@amd.com>

Author: volen-silo

crates/rocm-dash-tui/src/app/chat.rs

@@ -282,8 +282,28 @@ pub(super) fn config_with_chat(
     cfg
 }
 
+/// Fold a `/v1/models`-discovered model id into a resolved [`LlmConfig`].
+///
+/// Only overrides when the user did **not** configure a model explicitly (so the
+/// resolver filled in the neutral `DEFAULT_CHAT_MODEL` placeholder). An explicit
+/// `--chat-model`/config value always wins; a failed discovery (`None`) leaves
+/// the placeholder in place so endpoints that ignore the model field keep
+/// working. Pure — the unit-test anchor for the startup model-discovery wiring.
+pub(super) fn with_discovered_model(
+    mut llm: crate::llm::LlmConfig,
+    explicit_model: Option<&str>,
+    discovered: Option<String>,
+) -> crate::llm::LlmConfig {
+    if explicit_model.is_none()
+        && let Some(model) = discovered
+    {
+        llm.model = model;
+    }
+    llm
+}
+
 /// GET `{base}/models` and return the first served model id, or `None`.
-async fn fetch_first_model(base_url: &str) -> Option<String> {
+pub(super) async fn fetch_first_model(base_url: &str) -> Option<String> {
     let url = format!("{}/models", base_url.trim_end_matches('/'));
     let client = reqwest::Client::builder()
         .timeout(std::time::Duration::from_secs(3))
@@ -408,4 +428,36 @@ mod tests {
         let env = services_envelope(serde_json::json!([service("vllm", "", "ready", "m", 100)]));
         assert_eq!(pick_managed_chat_endpoint(&env), None);
     }
+
+    fn placeholder_config() -> crate::llm::LlmConfig {
+        crate::llm::detected_llm_config(
+            "http://127.0.0.1:11435/v1",
+            crate::llm::DEFAULT_CHAT_MODEL,
+        )
+    }
+
+    #[test]
+    fn discovered_model_replaces_placeholder_when_none_configured() {
+        // The 404 regression: no explicit model → the served id must be adopted.
+        let out = with_discovered_model(
+            placeholder_config(),
+            None,
+            Some("Qwen/Qwen2.5-1.5B-Instruct".to_string()),
+        );
+        assert_eq!(out.model, "Qwen/Qwen2.5-1.5B-Instruct");
+    }
+
+    #[test]
+    fn explicit_model_is_never_overridden() {
+        let mut cfg = placeholder_config();
+        cfg.model = "my-model".to_string();
+        let out = with_discovered_model(cfg, Some("my-model"), Some("served-id".to_string()));
+        assert_eq!(out.model, "my-model");
+    }
+
+    #[test]
+    fn failed_discovery_keeps_placeholder() {
+        let out = with_discovered_model(placeholder_config(), None, None);
+        assert_eq!(out.model, crate::llm::DEFAULT_CHAT_MODEL);
+    }
 }

crates/rocm-dash-tui/src/app/mod.rs

@@ -39,8 +39,8 @@ mod slash;
 mod summary;
 
 use chat::{
-    build_chat_agent, build_local_agent, detect_local_chat, detect_managed_chat,
-    persist_chat_endpoint,
+    build_chat_agent, build_local_agent, detect_local_chat, detect_managed_chat, fetch_first_model,
+    persist_chat_endpoint, with_discovered_model,
 };
 use summary::{parse_plan_result, summarize_json_value, summarize_slash_tool};
 
@@ -1582,6 +1582,9 @@ async fn event_loop(terminal: &mut Tui, args: &ResolvedArgs) -> color_eyre::Resu
             .await
             .unwrap_or(false)
         };
+        // The managed branch already carries a `/v1/models`-discovered model; a
+        // resolver-produced config from a configured URL does not.
+        let had_managed = managed.is_some();
         let llm = managed.or_else(|| {
             crate::llm::resolve_llm_config(
                 args.chat_url.as_deref(),
@@ -1594,6 +1597,22 @@ async fn event_loop(terminal: &mut Tui, args: &ResolvedArgs) -> color_eyre::Resu
                 probe_ok,
             )
         });
+        // A configured URL (env/CLI) with no explicit model resolves to the
+        // `local-model` placeholder, which 404s on servers that register the
+        // model under its real id. Discover the served model via `/v1/models`,
+        // mirroring the managed-service path. Gated on `probe_ok` so an
+        // unreachable endpoint never adds the fetch timeout to startup.
+        let llm = match llm {
+            Some(cfg) if !had_managed && probe_ok && args.chat_model.is_none() => {
+                let discovered = fetch_first_model(&cfg.base_url).await;
+                Some(with_discovered_model(
+                    cfg,
+                    args.chat_model.as_deref(),
+                    discovered,
+                ))
+            }
+            other => other,
+        };
         state.set_chat_config(llm, args.chat_auto_consent);
         // No reachable local endpoint AND no key/url configured → the no-key
         // ChatGPT OAuth default (device-code login surfaced in the chat tab).