From 100d6491f727a7e93bb433c3b950868a23785554 Mon Sep 17 00:00:00 2001
From: ciregenz <ciregenz@berkeley.edu>
Date: Wed, 10 Jun 2026 19:10:57 -0700
Subject: [PATCH] [eric] models: track model errors (diagnostic) + friendly
 'not on your plan' card for the 1211 unknown-model wall

---
 backend/apps/agents/agent_manager.py          | 36 +++++++++++++++++++
 backend/apps/agents/core/error_classify.py    | 20 +++++++++++
 .../pages/AgentChat/bubbles/MessageBubble.tsx | 10 ++++++
 3 files changed, 66 insertions(+)

diff --git a/backend/apps/agents/agent_manager.py b/backend/apps/agents/agent_manager.py
index 955ef29f..276478d2 100644
--- a/backend/apps/agents/agent_manager.py
+++ b/backend/apps/agents/agent_manager.py
@@ -32,6 +32,7 @@ from backend.apps.agents.core.error_classify import (
     _is_auth_error,
     _is_long_context_error,
     _is_transient_capacity_error,
+    _is_unknown_model_error,
 )
 from backend.apps.agents.manager.session.session_store import (
     _delete_session_file,
@@ -3009,7 +3010,42 @@ class AgentManager:
                     "session_id": session_id,
                     "message": error_msg.model_dump(mode="json"),
                 })
+            elif _is_unknown_model_error(e, extra_text=_stderr_tail):
+                # Upstream rejected the model code itself (e.g. Codex 1211 on a
+                # ChatGPT plan that lacks our GPT ids). Track it; the friendly
+                # "add an API key / pick another model" card is rendered frontend-side.
+                try:
+                    from backend.apps.service.client import submit_diagnostic
+                    submit_diagnostic({
+                        "kind": "model_error",
+                        "subkind": "unknown_model",
+                        "model": session.model,
+                        "provider": session.provider,
+                        "connection_mode": getattr(load_settings(), "connection_mode", "own_key"),
+                        "error_preview": (str(e) or "")[:400],
+                    })
+                except Exception:
+                    logger.debug("submit_diagnostic model_error failed", exc_info=True)
+                error_msg = Message(role="system", content=f"Error: {str(e)}", branch_id=session.active_branch_id)
+                session.messages.append(error_msg)
+                await ws_manager.send_to_session(session_id, "agent:message", {
+                    "session_id": session_id,
+                    "message": error_msg.model_dump(mode="json"),
+                })
             else:
+                # Track unclassified agent failures too so we stop flying blind on them.
+                try:
+                    from backend.apps.service.client import submit_diagnostic
+                    submit_diagnostic({
+                        "kind": "model_error",
+                        "subkind": "unclassified",
+                        "model": session.model,
+                        "provider": session.provider,
+                        "connection_mode": getattr(load_settings(), "connection_mode", "own_key"),
+                        "error_preview": (str(e) or "")[:400],
+                    })
+                except Exception:
+                    logger.debug("submit_diagnostic model_error failed", exc_info=True)
                 error_msg = Message(role="system", content=f"Error: {str(e)}", branch_id=session.active_branch_id)
                 session.messages.append(error_msg)
                 await ws_manager.send_to_session(session_id, "agent:message", {
diff --git a/backend/apps/agents/core/error_classify.py b/backend/apps/agents/core/error_classify.py
index 9add3ec3..b469e6da 100644
--- a/backend/apps/agents/core/error_classify.py
+++ b/backend/apps/agents/core/error_classify.py
@@ -77,6 +77,26 @@ def _is_auth_error(exc: BaseException, extra_text: str = "") -> bool:
     ))
 
 
+def _is_unknown_model_error(exc: BaseException, extra_text: str = "") -> bool:
+    """True when the upstream rejects the model code itself (e.g. a ChatGPT/Codex
+    subscription whose plan doesn't expose the GPT model id we send: code 1211
+    'Unknown Model, please check the model code'). The fix isn't retry, it's a
+    different model or an API key, so we surface that instead of the raw JSON.
+    """
+    combined = f"{exc!s}\n{extra_text}".strip()
+    if not combined:
+        return False
+    return bool(re.search(
+        r"unknown\s+model"
+        r"|check\s+the\s+model\s+code"
+        r"|\b1211\b"
+        r"|model[_\s-]?not[_\s-]?found"
+        r"|does\s+not\s+exist.*model|model.*does\s+not\s+exist",
+        combined,
+        re.IGNORECASE,
+    ))
+
+
 def _is_transient_capacity_error(exc: BaseException, extra_text: str = "") -> bool:
     # The Claude CLI's underlying ProcessError stringifies to a generic
     # "Command failed with exit code 1 / Check stderr output for details";
diff --git a/frontend/src/app/pages/AgentChat/bubbles/MessageBubble.tsx b/frontend/src/app/pages/AgentChat/bubbles/MessageBubble.tsx
index b9cb6f03..dcf08c30 100644
--- a/frontend/src/app/pages/AgentChat/bubbles/MessageBubble.tsx
+++ b/frontend/src/app/pages/AgentChat/bubbles/MessageBubble.tsx
@@ -101,6 +101,16 @@ function parseOpenSwarmError(text: string, ctx?: OverflowContext): OpenSwarmErro
       ctaAction: 'upgrade',
     };
   }
+  if (/unknown model|check the model code|\b1211\b|model_not_found/i.test(text)) {
+    return {
+      kind: 'auth',
+      title: "That model isn't available on your plan",
+      detail:
+        "Your connected subscription doesn't include this model. Add an API key for it in Settings, or pick a different model.",
+      ctaLabel: 'Open Settings',
+      ctaAction: 'settings',
+    };
+  }
   if (/at capacity|Try again shortly|503|service unavailable/i.test(text)) {
     return {
       kind: 'network',