feat(webui): show runtime model and fallback indicator in sidebar badge

agent/agent_init.py

@@ -287,7 +287,11 @@ def init_agent(
     """
     _install_safe_stdio()
 
-    agent.model = model
+    # Defensive: strip accidentally duplicated model-name prefix (see #54511).
+    from agent.chat_completion_helpers import _dedupe_model_name
+    agent.model = _dedupe_model_name(model)
+    import _sys as _sys_mod
+    print(f"[INIT_TRACE] entry: model={model!r} deduped={agent.model!r} provider={provider!r} base_url={base_url!r}", file=_sys_mod.stderr, flush=True)
     agent.max_iterations = max_iterations
     # Shared iteration budget — parent creates, children inherit.
     # Consumed by every LLM turn across parent + all subagents.

agent/agent_runtime_helpers.py

@@ -1109,6 +1109,8 @@ def restore_primary_runtime(agent) -> bool:
         # ── Reset fallback chain for the new turn ──
         agent._fallback_activated = False
         agent._fallback_index = 0
+        agent._primary_model = ""
+        agent._primary_provider = ""
 
         # Undo the fallback's identity rewrite so the prompt is
         # byte-identical to the stored copy again (prefix cache match).
@@ -1580,7 +1582,11 @@ def switch_model(agent, new_model, new_provider, api_key='', base_url='', api_mo
         agent._config_context_length = None
 
         # ── Swap core runtime fields ──
-        agent.model = new_model
+        # Defensive: strip accidentally duplicated model-name prefix
+        # (see #54511).  The /model command can produce doubled names like
+        # "gemma4:gemma4:e2b" from alias resolution; dedupe before storing.
+        from agent.chat_completion_helpers import _dedupe_model_name
+        agent.model = _dedupe_model_name(new_model)
         agent.provider = new_provider
         # Use new base_url when provided; only fall back to current when the
         # new provider genuinely has no endpoint (e.g. native SDK providers).
@@ -1774,6 +1780,8 @@ def switch_model(agent, new_model, new_provider, api_key='', base_url='', api_mo
     # ── Reset fallback state ──
     agent._fallback_activated = False
     agent._fallback_index = 0
+    agent._primary_model = ""
+    agent._primary_provider = ""
 
     # When the user deliberately swaps primary providers (e.g. openrouter
     # → anthropic), drop any fallback entries that target the OLD primary

agent/chat_completion_helpers.py

@@ -586,9 +586,40 @@ def _call():
 
 
 
+def _dedupe_model_name(name: str) -> str:
+    """Strip accidentally duplicated provider/model prefixes.
+
+    Ollama model tags like ``gemma4:e2b`` can get doubled to
+    ``gemma4:gemma4:e2b`` by an upstream bug (#54511).  A doubled tag is
+    an invalid model name and the API call fails with HTTP 400.
+
+    Handles one or more duplications: ``a:a:a:b`` → ``a:b``.
+    """
+    if not name or ":" not in name:
+        return name
+    parts = name.split(":")
+    # Walk from the left, collapsing runs where part[i] == part[i+1].
+    deduped = [parts[0]]
+    for p in parts[1:]:
+        if p != deduped[-1]:
+            deduped.append(p)
+    # If the result still has the pattern ``prefix:prefix:model`` after
+    # adjacent-collapse (e.g. ``gemma4:gemma4:e2b``), also strip a leading
+    # duplicate of the final segment's prefix.
+    result = ":".join(deduped)
+    if ":" in result:
+        head, _, tail = result.rpartition(":")
+        if head.endswith(":" + tail) or head == tail:
+            result = tail
+    return result
+
+
 def build_api_kwargs(agent, api_messages: list) -> dict:
     """Build the keyword arguments dict for the active API mode."""
     tools_for_api = agent.tools
+    # Defensive: strip any accidentally duplicated model-name prefix
+    # (see #54511).  This is a no-op for well-formed names.
+    agent.model = _dedupe_model_name(agent.model)
 
     if agent.api_mode == "anthropic_messages":
         _transport = agent._get_transport()
@@ -1247,6 +1278,13 @@ def try_activate_fallback(agent, reason: "FailoverReason | None" = None) -> bool
 
         old_model = agent.model
 
+        # Remember the primary (original) model/provider so the UI can show
+        # "Fallback: X (primary: Y)" — stored once, before the swap.
+        if not getattr(agent, "_primary_model", ""):
+            agent._primary_model = getattr(agent, "model", "") or ""
+        if not getattr(agent, "_primary_provider", ""):
+            agent._primary_provider = getattr(agent, "provider", "") or ""
+
         # Clear the per-config context_length override so the fallback
         # model's actual context window is resolved instead of inheriting
         # the stale value from the previous model.  See #22387.
@@ -2190,6 +2228,7 @@ def _call_chat_completions():
                 content=full_content,
                 tool_calls=None,
                 reasoning_content=full_reasoning,
+                reasoning=full_reasoning,
             )
             mock_choice = SimpleNamespace(
                 index=0,
@@ -2214,6 +2253,7 @@ def _call_chat_completions():
             content=full_content,
             tool_calls=mock_tool_calls,
             reasoning_content=full_reasoning,
+            reasoning=full_reasoning,
         )
         mock_choice = SimpleNamespace(
             index=0,

agent/conversation_loop.py

@@ -4361,6 +4361,13 @@ def _perform_api_call(next_api_kwargs):
             else:
                 # No tool calls - this is the final response
                 final_response = assistant_message.content or ""
+                # Ollama + gemma4 puts vision/multimodal output in `reasoning`
+                # instead of `content`. Fall back to reasoning when content is
+                # empty but reasoning carries the actual answer.
+                if not final_response.strip():
+                    _ollama_reasoning = getattr(assistant_message, "reasoning", None)
+                    if isinstance(_ollama_reasoning, str) and _ollama_reasoning.strip():
+                        final_response = _ollama_reasoning.strip()
 
                 # Fix: unmute output when entering the no-tool-call branch
                 # so the user can see empty-response warnings and recovery

agent/model_metadata.py

@@ -1837,13 +1837,24 @@ def get_model_context_length(
                 key=lambda x: len(x[0]),
                 reverse=True,
             ):
-                if default_model in model_lower:
-                    logger.info(
-                        "Using hardcoded context length %s for model %r "
-                        "(custom endpoint, catalog match on %r)",
-                        f"{length:,}", model, default_model,
-                    )
-                    return length
+                dm_lower = default_model.lower()
+                idx = model_lower.find(dm_lower)
+                if idx == -1:
+                    continue
+                # Reject mid-token matches: "gemma" inside "gemma4" is not
+                # a real match.  The char after the catalog key must be a
+                # delimiter (colon, dash, dot, etc.) or end-of-string.
+                end = idx + len(dm_lower)
+                if end < len(model_lower):
+                    next_ch = model_lower[end]
+                    if next_ch.isalnum() or next_ch == "_":
+                        continue
+                logger.info(
+                    "Using hardcoded context length %s for model %r "
+                    "(custom endpoint, catalog match on %r)",
+                    f"{length:,}", model, default_model,
+                )
+                return length
             return DEFAULT_FALLBACK_CONTEXT
 
     # 4. Anthropic /v1/models API (only for regular API keys, not OAuth)

gateway/run.py

@@ -3469,6 +3469,24 @@ def _resolve_session_agent_runtime(
         override = self._session_model_overrides.get(resolved_session_key) if resolved_session_key else None
         if override:
             override_model = override.get("model", model)
+            # Defense-in-depth: validate override model name before use.
+            # A corrupted override (doubled prefix, truncated tag) poisons
+            # every subsequent rebuild until /new.  Dedupe silently so a
+            # bad override still falls through to config rather than
+            # silently breaking the session.  See #54511.
+            try:
+                from agent.chat_completion_helpers import _dedupe_model_name
+                _validated = _dedupe_model_name(override_model)
+                if _validated != override_model:
+                    logger.warning(
+                        "Session override model name corrected: %r -> %r (session=%s)",
+                        override_model, _validated, resolved_session_key,
+                    )
+                    override_model = _validated
+                    # Patch the override in-place so subsequent turns are clean
+                    override["model"] = _validated
+            except Exception:
+                pass
             override_runtime = {
                 "provider": override.get("provider"),
                 "api_key": override.get("api_key"),
@@ -16026,8 +16044,8 @@ def run_sync():
                     session_key=session_key,
                     user_config=user_config,
                 )
-                logger.debug(
-                    "run_agent resolved: model=%s provider=%s session=%s",
+                logger.warning(
+                    "DBG_RESOLVED model=%s provider=%s session=%s",
                     model, runtime_kwargs.get("provider"), session_key or "",
                 )
             except Exception as exc:
@@ -16267,6 +16285,7 @@ def _interim_assistant_cb(text: str, *, already_streamed: bool = False) -> None:
 
             if agent is None:
                 # Config changed or first message — create fresh agent
+                logger.warning("DBG_AGENT_CREATE model=%s sig=%s cached=%s", turn_route["model"], _sig, reused_cached_agent)
                 agent = AIAgent(
                     model=turn_route["model"],
                     **turn_route["runtime"],

gateway/slash_commands.py

@@ -1557,6 +1557,24 @@ async def _finish_switch() -> str:
                 f"Adjust your self-identification accordingly.]"
             )
 
+            # Validate model name before storing override — a corrupted name
+            # (doubled prefix like "gemma4:gemma4:e2b" or truncated like "e2b")
+            # would poison every subsequent turn until /new or manual DB edit.
+            # See #54511.
+            _validated_model = result.new_model
+            try:
+                from agent.chat_completion_helpers import _dedupe_model_name
+                _validated_model = _dedupe_model_name(result.new_model)
+            except Exception:
+                pass  # dedupe is best-effort; fall back to raw name
+            if _validated_model != result.new_model:
+                logger.warning(
+                    "Session override model name corrected: %r -> %r",
+                    result.new_model,
+                    _validated_model,
+                )
+                result.new_model = _validated_model
+
             # Store session override so next agent creation uses the new model
             self._session_model_overrides[session_key] = {
                 "model": result.new_model,

run_agent.py

@@ -528,6 +528,7 @@ def _ensure_db_session(self) -> None:
             return
         source = _session_source_for_agent(self.platform)
         try:
+            logger.warning("DBG_DB_CREATE session=%s model=%s provider=%s", self.session_id, self.model, self.provider)
             self._session_db.create_session(
                 session_id=self.session_id,
                 source=source,

tui_gateway/server.py

@@ -3085,6 +3085,9 @@ def _session_info(agent, session: dict | None = None) -> dict:
     info: dict = {
         "model": getattr(agent, "model", ""),
         "provider": getattr(agent, "provider", ""),
+        "fallback_activated": bool(getattr(agent, "_fallback_activated", False)),
+        "primary_model": getattr(agent, "_primary_model", "") or "",
+        "primary_provider": getattr(agent, "_primary_provider", "") or "",
         "reasoning_effort": reasoning_effort,
         "service_tier": service_tier,
         "fast": service_tier == "priority",

web/src/components/ChatSidebar.tsx

@@ -42,6 +42,9 @@ interface SessionInfo {
   cwd?: string;
   model?: string;
   provider?: string;
+  fallback_activated?: boolean;
+  primary_model?: string;
+  primary_provider?: string;
   credential_warning?: string;
   title?: string;
 }
@@ -303,8 +306,21 @@ export function ChatSidebar({
 
   // The picker writes config.yaml over REST and reloads — it doesn't ride the
   // sidecar gateway session, so it's available whenever the sidebar is mounted.
-  const modelName = effectiveModel || info.model || "—";
+  //
+  // When fallback is active, prefer the runtime model from the gateway WS
+  // (info.model) over the config-API value (effectiveModel) so the badge
+  // reflects what's actually generating responses.
+  const fallbackActive = !!info.fallback_activated;
+  const modelName = fallbackActive
+    ? info.model || effectiveModel || "—"
+    : effectiveModel || info.model || "—";
   const modelLabel = modelName.split("/").slice(-1)[0] ?? "—";
+  const primaryLabel =
+    (info.primary_model || info.primary_provider) &&
+    fallbackActive &&
+    info.model !== info.primary_model
+      ? `${(info.primary_model || "").split("/").slice(-1)[0] || info.primary_provider || ""}`
+      : null;
   const banner = error ?? info.credential_warning ?? null;
 
   return (
@@ -320,23 +336,44 @@ export function ChatSidebar({
             model
           </div>
 
-          <Button
-            ghost
-            size="sm"
-            onClick={() => setModelOpen(true)}
-            className={cn(
-              "max-w-full min-w-0 px-0 py-0",
-              "self-start normal-case tracking-normal text-sm font-medium",
-              "hover:underline disabled:no-underline",
+          <div className="flex items-center gap-1.5">
+            <Button
+              ghost
+              size="sm"
+              onClick={() => setModelOpen(true)}
+              className={cn(
+                "max-w-full min-w-0 px-0 py-0",
+                "self-start normal-case tracking-normal text-sm font-medium",
+                "hover:underline disabled:no-underline",
+              )}
+              title={
+                fallbackActive && primaryLabel
+                  ? `Fallback: ${modelName} (primary: ${info.primary_model || info.primary_provider || "—"})`
+                  : modelName === "—"
+                    ? "switch model"
+                    : modelName
+              }
+            >
+              <span className="flex min-w-0 max-w-full items-center gap-1">
+                <span className="truncate">{modelLabel}</span>
+
+                <ChevronDown className="size-3.5 shrink-0 text-text-secondary" />
+              </span>
+            </Button>
+
+            {fallbackActive && (
+              <span
+                className="shrink-0 rounded-full bg-amber-500/15 px-1.5 py-0.5 text-[10px] font-semibold uppercase tracking-wide text-amber-600 dark:text-amber-400"
+                title={
+                  primaryLabel
+                    ? `Fallback active — primary ${info.primary_model || info.primary_provider || "—"} unavailable`
+                    : "Fallback active"
+                }
+              >
+                Fallback
+              </span>
             )}
-            title={modelName === "—" ? "switch model" : modelName}
-          >
-            <span className="flex min-w-0 max-w-full items-center gap-1">
-              <span className="truncate">{modelLabel}</span>
-
-              <ChevronDown className="size-3.5 shrink-0 text-text-secondary" />
-            </span>
-          </Button>
+          </div>
         </div>
 
         <Badge tone={STATE_TONE[state]} className="shrink-0">