Skip to content

The agent that grows with you

fix: a down local model server reads as itself in the TUI, dashboard and chat replies

Teknium6c625765d69e
diff --git a/hermes_cli/runtime_provider_custom.py b/hermes_cli/runtime_provider_custom.py
index 6b8439ed942..72913cf22a2 100644
--- a/hermes_cli/runtime_provider_custom.py
+++ b/hermes_cli/runtime_provider_custom.py
@@ -617,11 +617,12 @@ def _resolve_llamacpp_runtime(requested_provider: str, explicit_api_key: Optiona
enabled = False
if enabled:
raise LocalRuntimeUnavailableError("The local model server isn't running. It may still be "
- "starting — try again in a moment, or check Settings → "
- "Providers → Local models.", retryable=True)
- raise LocalRuntimeUnavailableError("The local model server is turned off. Turn it back on in "
- "Settings → Providers → Local models, or switch to another "
- "model.")
+ "starting — try again in a moment, or switch to another "
+ "model with /model.", retryable=True)
+ # Every chat surface (messaging, TUI, Webapp) shows this sentence, so it names no Desktop-only
+ # pane; the Desktop client links its Local models pane itself.
+ raise LocalRuntimeUnavailableError("The local model server is turned off. Turn it back on in the "
+ "Hermes desktop app, or switch to another model with /model.")
def _custom_runtime(rp, base_url: str, api_key: Any, api_mode: Optional[str], **extra: Any) -> dict[str, Any]:
diff --git a/tests/hermes_cli/test_local_runtime_unavailable_surface.py b/tests/hermes_cli/test_local_runtime_unavailable_surface.py
index ec905f7040c..e3c543845e9 100644
--- a/tests/hermes_cli/test_local_runtime_unavailable_surface.py
+++ b/tests/hermes_cli/test_local_runtime_unavailable_surface.py
@@ -74,3 +74,29 @@ def test_turn_failure_is_an_endpoint_failure_not_a_rejected_key(llamacpp_home, e
assert surface["layer"] == "endpoint"
assert surface["code"] == "local_runtime_unavailable"
assert "auth_kind" not in surface
+
+
[email protected]("enabled", [False, True])
+def test_chat_text_names_the_local_server_and_no_desktop_only_pane(llamacpp_home, enabled):
+ """The messaging reply and the TUI/dashboard turn text both name the local model server, and
+ neither points at Settings → Providers → Local models, a pane only the Desktop app has."""
+ from agent.error_surface import build_error_surface_from_result
+ from gateway.config import Platform
+ from gateway.run_turn_runner import _resolution_failure_reply
+ from tui_gateway.user_messages import turn_error_text
+
+ llamacpp_home({"enabled": enabled})
+ exc = _resolution_error()
+ reply, reason, retryable = _resolution_failure_reply(exc, Platform.TELEGRAM)
+ surface = build_error_surface_from_result(
+ {"failed": True, "error": str(exc), "failure_reason": reason, "failure_retryable": retryable},
+ provider="llamacpp")
+ tui = turn_error_text(str(exc), surface, recoverable=retryable)
+
+ title = tui.splitlines()[0]
+ assert "local model server" in title
+ assert "custom model endpoint" not in tui
+ for text in (reply, tui):
+ assert "local model server" in text
+ assert "Settings" not in text and "Local models" not in text
+ assert "/model" in text
diff --git a/tui_gateway/user_messages.py b/tui_gateway/user_messages.py
index 6e73065e996..6b58b909cd5 100644
--- a/tui_gateway/user_messages.py
+++ b/tui_gateway/user_messages.py
@@ -31,6 +31,9 @@ _TURN_ERROR_CODE_COPY: dict[str, tuple[str, str]] = {
"content_policy_blocked": ("The model provider refused this request (content policy)", "Rephrase and send again."),
"provider_policy_blocked": ("The model provider refused this request (account policy)", "Switch with /model."),
"format_error": ("The model provider rejected the request format", "Try /retry; if it persists, switch with /model."),
+ # The managed local model server (llama.cpp) is down: starting, or switched off.
+ "local_runtime_unavailable": ("The local model server is not running",
+ "Try /retry once it is up, or switch with /model."),
"ssl_cert_verification": ("The connection to the model provider could not be verified (TLS)",
"Check the endpoint's certificate, then /retry."),
}
diff --git a/ui-tui/src/__tests__/userMessages.test.ts b/ui-tui/src/__tests__/userMessages.test.ts
index ccde8dae8b4..613b275f3e7 100644
--- a/ui-tui/src/__tests__/userMessages.test.ts
+++ b/ui-tui/src/__tests__/userMessages.test.ts
@@ -43,6 +43,17 @@ describe('describeTurnFailure', () => {
expect(text).toContain('/retry')
})
+ it('names the local model server when the managed local runtime is down, not a custom endpoint', () => {
+ const text = describeTurnFailure({
+ error: "The local model server isn't running. It may still be starting.",
+ error_surface: { code: 'local_runtime_unavailable', layer: 'endpoint', provider: 'llamacpp', retryable: true }
+ })
+
+ expect(text.split('\n')[0]).toContain('local model server')
+ expect(text).not.toContain('custom model endpoint')
+ expect(text).toContain('/model')
+ })
+
it('falls back to the layer copy, and to a generic title, when the code is unknown', () => {
const streaming = describeTurnFailure({
error: 'peer closed connection',
diff --git a/ui-tui/src/app/userMessages.ts b/ui-tui/src/app/userMessages.ts
index f63c7649162..8ad93d9d34a 100644
--- a/ui-tui/src/app/userMessages.ts
+++ b/ui-tui/src/app/userMessages.ts
@@ -271,6 +271,7 @@ const TURN_CODE_KEY: Record<string, keyof Translations['userMessages']['turn']['
content_policy_blocked: 'contentPolicyBlocked',
context_overflow: 'contextOverflow',
format_error: 'formatError',
+ local_runtime_unavailable: 'localRuntimeUnavailable',
model_not_found: 'modelNotFound',
overloaded: 'overloaded',
payload_too_large: 'payloadTooLarge',
diff --git a/ui-tui/src/i18n/en/userMessages.ts b/ui-tui/src/i18n/en/userMessages.ts
index 06ae4774d26..99e029416ef 100644
--- a/ui-tui/src/i18n/en/userMessages.ts
+++ b/ui-tui/src/i18n/en/userMessages.ts
@@ -100,6 +100,11 @@ export const userMessagesEn = {
hint: 'Try /retry; if it persists, switch with /model.',
hintNoRetry: 'Pick another model with /model; if it persists, switch with /model.'
},
+ localRuntimeUnavailable: {
+ title: 'The local model server is not running',
+ hint: 'Try /retry once it is up, or switch with /model.',
+ hintNoRetry: 'Turn it back on, or switch with /model.'
+ },
modelNotFound: {
title: 'The model provider does not know this model',
hint: 'Pick another model with /model.'