fix(copilot): handle transient Anthropic API connection errors gracefully (#12445)

## Summary
- Detect transient Anthropic API errors (ECONNRESET, "socket connection
was closed unexpectedly") across all error paths in the copilot SDK
streaming loop
- Replace raw technical error messages with user-friendly text:
**"Anthropic connection interrupted — please retry"**
- Add `retryable` field to `StreamError` model so the frontend can
distinguish retryable errors
- Add **"Try Again" button** on the error card for transient errors,
which re-sends the last user message

### Background
Sentry issue
[AUTOGPT-SERVER-875](https://significant-gravitas.sentry.io/issues/AUTOGPT-SERVER-875)
— 25+ events since March 13, caused by Anthropic API infrastructure
instability (confirmed by their status page). Same SDK/code on dev and
prod, prod-only because of higher volume of long-running streaming
sessions.

### Changes
**Backend (`constants.py`, `service.py`, `response_adapter.py`,
`response_model.py`):**
- `is_transient_api_error()` — pattern-matching helper for known
transient error strings
- Intercept transient errors in 3 places: `AssistantMessage.error`,
stream exceptions, `BaseException` handler
- Use friendly message in error markers persisted to session (so it
shows properly on page refresh too)
- `StreamError.retryable` field for frontend consumption

**Frontend (`ChatContainer`, `ChatMessagesContainer`,
`MessagePartRenderer`):**
- Thread `onRetry` callback from `ChatContainer` →
`ChatMessagesContainer` → `MessagePartRenderer`
- Detect transient error text in error markers and show "Try Again"
button via existing `ErrorCard.onRetry`
- Clicking "Try Again" re-sends the last user message (backend
auto-cleans stale error markers)

Fixes SECRT-2128, SECRT-2129, SECRT-2130

## Test plan
- [ ] Verify transient error detection with `is_transient_api_error()`
for known patterns
- [ ] Confirm error card shows "Anthropic connection interrupted —
please retry" instead of raw socket error
- [ ] Confirm "Try Again" button appears on transient error cards
- [ ] Confirm "Try Again" re-sends the last user message successfully
- [ ] Confirm non-transient errors (e.g., "Prompt is too long") still
show original error text without retry button
- [ ] Verify error marker persists correctly on page refresh

autogpt_platform/backend/backend/copilot/config.py

@@ -115,10 +115,22 @@ class ChatConfig(BaseSettings):
         description="Use --resume for multi-turn conversations instead of "
         "history compression. Falls back to compression when unavailable.",
     )
+    use_openrouter: bool = Field(
+        default=True,
+        description="Enable routing API calls through the OpenRouter proxy. "
+        "The actual decision also requires ``api_key`` and ``base_url`` — "
+        "use the ``openrouter_active`` property for the final answer.",
+    )
     use_claude_code_subscription: bool = Field(
         default=False,
         description="For personal/dev use: use Claude Code CLI subscription auth instead of API keys. Requires `claude login` on the host. Only works with SDK mode.",
     )
+    test_mode: bool = Field(
+        default=False,
+        description="Use dummy service instead of real LLM calls. "
+        "Send __test_transient_error__, __test_fatal_error__, or "
+        "__test_slow_response__ to trigger specific scenarios.",
+    )
 
     # E2B Sandbox Configuration
     use_e2b_sandbox: bool = Field(
@@ -146,6 +158,21 @@ class ChatConfig(BaseSettings):
         description="E2B lifecycle action on timeout: 'pause' (default, free) or 'kill'.",
     )
 
+    @property
+    def openrouter_active(self) -> bool:
+        """True when OpenRouter is enabled AND credentials are usable.
+
+        Single source of truth for "will the SDK route through OpenRouter?".
+        Checks the flag *and* that ``api_key`` + a valid ``base_url`` are
+        present — mirrors the fallback logic in ``_build_sdk_env``.
+        """
+        if not self.use_openrouter:
+            return False
+        base = (self.base_url or "").rstrip("/")
+        if base.endswith("/v1"):
+            base = base[:-3]
+        return bool(self.api_key and base and base.startswith("http"))
+
     @property
     def e2b_active(self) -> bool:
         """True when E2B is enabled and the API key is present.
@@ -168,15 +195,6 @@ class ChatConfig(BaseSettings):
         """
         return self.e2b_api_key if self.e2b_active else None
 
-    @field_validator("use_e2b_sandbox", mode="before")
-    @classmethod
-    def get_use_e2b_sandbox(cls, v):
-        """Get use_e2b_sandbox from environment if not provided."""
-        env_val = os.getenv("CHAT_USE_E2B_SANDBOX", "").lower()
-        if env_val:
-            return env_val in ("true", "1", "yes", "on")
-        return True if v is None else v
-
     @field_validator("e2b_api_key", mode="before")
     @classmethod
     def get_e2b_api_key(cls, v):
@@ -219,26 +237,6 @@ class ChatConfig(BaseSettings):
                 v = OPENROUTER_BASE_URL
         return v
 
-    @field_validator("use_claude_agent_sdk", mode="before")
-    @classmethod
-    def get_use_claude_agent_sdk(cls, v):
-        """Get use_claude_agent_sdk from environment if not provided."""
-        # Check environment variable - default to True if not set
-        env_val = os.getenv("CHAT_USE_CLAUDE_AGENT_SDK", "").lower()
-        if env_val:
-            return env_val in ("true", "1", "yes", "on")
-        # Default to True (SDK enabled by default)
-        return True if v is None else v
-
-    @field_validator("use_claude_code_subscription", mode="before")
-    @classmethod
-    def get_use_claude_code_subscription(cls, v):
-        """Get use_claude_code_subscription from environment if not provided."""
-        env_val = os.getenv("CHAT_USE_CLAUDE_CODE_SUBSCRIPTION", "").lower()
-        if env_val:
-            return env_val in ("true", "1", "yes", "on")
-        return False if v is None else v
-
     # Prompt paths for different contexts
     PROMPT_PATHS: dict[str, str] = {
         "default": "prompts/chat_system.md",
@@ -248,6 +246,7 @@ class ChatConfig(BaseSettings):
     class Config:
         """Pydantic config."""
 
+        env_prefix = "CHAT_"
         env_file = ".env"
         env_file_encoding = "utf-8"
         extra = "ignore"  # Ignore extra environment variables

autogpt_platform/backend/backend/copilot/config_test.py

@@ -6,19 +6,70 @@ from .config import ChatConfig
 
 # Env vars that the ChatConfig validators read — must be cleared so they don't
 # override the explicit constructor values we pass in each test.
-_E2B_ENV_VARS = (
+_ENV_VARS_TO_CLEAR = (
     "CHAT_USE_E2B_SANDBOX",
     "CHAT_E2B_API_KEY",
     "E2B_API_KEY",
+    "CHAT_USE_OPENROUTER",
+    "CHAT_API_KEY",
+    "OPEN_ROUTER_API_KEY",
+    "OPENAI_API_KEY",
+    "CHAT_BASE_URL",
+    "OPENROUTER_BASE_URL",
+    "OPENAI_BASE_URL",
 )
 
 
 @pytest.fixture(autouse=True)
-def _clean_e2b_env(monkeypatch: pytest.MonkeyPatch) -> None:
-    for var in _E2B_ENV_VARS:
+def _clean_env(monkeypatch: pytest.MonkeyPatch) -> None:
+    for var in _ENV_VARS_TO_CLEAR:
         monkeypatch.delenv(var, raising=False)
 
 
+class TestOpenrouterActive:
+    """Tests for the openrouter_active property."""
+
+    def test_enabled_with_credentials_returns_true(self):
+        cfg = ChatConfig(
+            use_openrouter=True,
+            api_key="or-key",
+            base_url="https://openrouter.ai/api/v1",
+        )
+        assert cfg.openrouter_active is True
+
+    def test_enabled_but_missing_api_key_returns_false(self):
+        cfg = ChatConfig(
+            use_openrouter=True,
+            api_key=None,
+            base_url="https://openrouter.ai/api/v1",
+        )
+        assert cfg.openrouter_active is False
+
+    def test_disabled_returns_false_despite_credentials(self):
+        cfg = ChatConfig(
+            use_openrouter=False,
+            api_key="or-key",
+            base_url="https://openrouter.ai/api/v1",
+        )
+        assert cfg.openrouter_active is False
+
+    def test_strips_v1_suffix_and_still_valid(self):
+        cfg = ChatConfig(
+            use_openrouter=True,
+            api_key="or-key",
+            base_url="https://openrouter.ai/api/v1",
+        )
+        assert cfg.openrouter_active is True
+
+    def test_invalid_base_url_returns_false(self):
+        cfg = ChatConfig(
+            use_openrouter=True,
+            api_key="or-key",
+            base_url="not-a-url",
+        )
+        assert cfg.openrouter_active is False
+
+
 class TestE2BActive:
     """Tests for the e2b_active property — single source of truth for E2B usage."""
 

autogpt_platform/backend/backend/copilot/constants.py

@@ -4,6 +4,9 @@
 # The hex suffix makes accidental LLM generation of these strings virtually
 # impossible, avoiding false-positive marker detection in normal conversation.
 COPILOT_ERROR_PREFIX = "[__COPILOT_ERROR_f7a1__]"  # Renders as ErrorCard
+COPILOT_RETRYABLE_ERROR_PREFIX = (
+    "[__COPILOT_RETRYABLE_ERROR_a9c2__]"  # ErrorCard + retry
+)
 COPILOT_SYSTEM_PREFIX = "[__COPILOT_SYSTEM_e3b0__]"  # Renders as system info message
 
 # Prefix for all synthetic IDs generated by CoPilot block execution.
@@ -35,3 +38,24 @@ def parse_node_id_from_exec_id(node_exec_id: str) -> str:
     Format: "{node_id}:{random_hex}" → returns "{node_id}".
     """
     return node_exec_id.rsplit(COPILOT_NODE_EXEC_ID_SEPARATOR, 1)[0]
+
+
+# ---------------------------------------------------------------------------
+# Transient Anthropic API error detection
+# ---------------------------------------------------------------------------
+# Patterns in error text that indicate a transient Anthropic API error
+# (ECONNRESET / dropped TCP connection) which is retryable.
+_TRANSIENT_ERROR_PATTERNS = (
+    "socket connection was closed unexpectedly",
+    "ECONNRESET",
+    "connection was forcibly closed",
+    "network socket disconnected",
+)
+
+FRIENDLY_TRANSIENT_MSG = "Anthropic connection interrupted — please retry"
+
+
+def is_transient_api_error(error_text: str) -> bool:
+    """Return True if *error_text* matches a known transient Anthropic API error."""
+    lower = error_text.lower()
+    return any(pat.lower() in lower for pat in _TRANSIENT_ERROR_PATTERNS)

autogpt_platform/backend/backend/copilot/executor/processor.py

@@ -16,6 +16,7 @@ from backend.copilot.baseline import stream_chat_completion_baseline
 from backend.copilot.config import ChatConfig
 from backend.copilot.response_model import StreamFinish
 from backend.copilot.sdk import service as sdk_service
+from backend.copilot.sdk.dummy import stream_chat_completion_dummy
 from backend.executor.cluster_lock import ClusterLock
 from backend.util.decorator import error_logged
 from backend.util.feature_flag import Flag, is_feature_enabled
@@ -246,17 +247,25 @@ class CoPilotProcessor:
             # Choose service based on LaunchDarkly flag.
             # Claude Code subscription forces SDK mode (CLI subprocess auth).
             config = ChatConfig()
-            use_sdk = config.use_claude_code_subscription or await is_feature_enabled(
-                Flag.COPILOT_SDK,
-                entry.user_id or "anonymous",
-                default=config.use_claude_agent_sdk,
-            )
-            stream_fn = (
-                sdk_service.stream_chat_completion_sdk
-                if use_sdk
-                else stream_chat_completion_baseline
-            )
-            log.info(f"Using {'SDK' if use_sdk else 'baseline'} service")
+
+            if config.test_mode:
+                stream_fn = stream_chat_completion_dummy
+                log.warning("Using DUMMY service (CHAT_TEST_MODE=true)")
+            else:
+                use_sdk = (
+                    config.use_claude_code_subscription
+                    or await is_feature_enabled(
+                        Flag.COPILOT_SDK,
+                        entry.user_id or "anonymous",
+                        default=config.use_claude_agent_sdk,
+                    )
+                )
+                stream_fn = (
+                    sdk_service.stream_chat_completion_sdk
+                    if use_sdk
+                    else stream_chat_completion_baseline
+                )
+                log.info(f"Using {'SDK' if use_sdk else 'baseline'} service")
 
             # Stream chat completion and publish chunks to Redis.
             async for chunk in stream_fn(

autogpt_platform/backend/backend/copilot/sdk/dummy.py

@@ -1,9 +1,17 @@
 """Dummy SDK service for testing copilot streaming.
 
 Returns mock streaming responses without calling Claude Agent SDK.
-Enable via COPILOT_TEST_MODE=true environment variable.
+Enable via CHAT_TEST_MODE=true in .env (ChatConfig.test_mode).
 
 WARNING: This is for testing only. Do not use in production.
+
+Magic keywords (case-insensitive, anywhere in message):
+    __test_transient_error__   — Simulate a transient Anthropic API error
+                                 (ECONNRESET).  Streams partial text, then
+                                 yields StreamError with retryable prefix.
+    __test_fatal_error__       — Simulate a non-retryable SDK error.
+    __test_slow_response__     — Simulate a slow response (2s per word).
+    (no keyword)               — Normal dummy response.
 """
 
 import asyncio
@@ -12,12 +20,39 @@ import uuid
 from collections.abc import AsyncGenerator
 from typing import Any
 
-from ..model import ChatSession
-from ..response_model import StreamBaseResponse, StreamStart, StreamTextDelta
+from ..constants import (
+    COPILOT_ERROR_PREFIX,
+    COPILOT_RETRYABLE_ERROR_PREFIX,
+    FRIENDLY_TRANSIENT_MSG,
+)
+from ..model import ChatMessage, ChatSession, get_chat_session, upsert_chat_session
+from ..response_model import (
+    StreamBaseResponse,
+    StreamError,
+    StreamFinish,
+    StreamFinishStep,
+    StreamStart,
+    StreamStartStep,
+    StreamTextDelta,
+    StreamTextEnd,
+    StreamTextStart,
+)
 
 logger = logging.getLogger(__name__)
 
 
+async def _safe_upsert(session: ChatSession) -> None:
+    """Best-effort session persist — skip silently if DB is unavailable."""
+    try:
+        await upsert_chat_session(session)
+    except Exception:
+        logger.debug("[TEST MODE] Could not persist session (DB unavailable)")
+
+
+def _has_keyword(message: str | None, keyword: str) -> bool:
+    return keyword in (message or "").lower()
+
+
 async def stream_chat_completion_dummy(
     session_id: str,
     message: str | None = None,
@@ -36,24 +71,89 @@ async def stream_chat_completion_dummy(
     - No timeout occurs
     - Text arrives in chunks
     - StreamFinish is sent by mark_session_completed
+
+    See module docstring for magic keywords that trigger error scenarios.
     """
     logger.warning(
         f"[TEST MODE] Using dummy copilot streaming for session {session_id}"
     )
 
+    # Load session from DB (matches SDK service behaviour) so error markers
+    # and the assistant reply are persisted and survive page refresh.
+    # Best-effort: skip if DB is unavailable (e.g. unit tests).
+    if session is None:
+        try:
+            session = await get_chat_session(session_id, user_id)
+        except Exception:
+            logger.debug("[TEST MODE] Could not load session (DB unavailable)")
+            session = None
+
     message_id = str(uuid.uuid4())
     text_block_id = str(uuid.uuid4())
 
-    # Start the stream
+    # Start the stream (matches baseline: StreamStart → StreamStartStep)
     yield StreamStart(messageId=message_id, sessionId=session_id)
+    yield StreamStartStep()
 
-    # Simulate streaming text response with delays
+    # --- Magic keyword: transient error (retryable) -------------------------
+    if _has_keyword(message, "__test_transient_error__"):
+        # Stream some partial text first (simulates mid-stream failure)
+        yield StreamTextStart(id=text_block_id)
+        for word in ["Working", "on", "it..."]:
+            yield StreamTextDelta(id=text_block_id, delta=f"{word} ")
+            await asyncio.sleep(0.1)
+        yield StreamTextEnd(id=text_block_id)
+        yield StreamFinishStep()
+        # Persist retryable marker so "Try Again" button shows after refresh
+        if session:
+            session.messages.append(
+                ChatMessage(
+                    role="assistant",
+                    content=f"{COPILOT_RETRYABLE_ERROR_PREFIX} {FRIENDLY_TRANSIENT_MSG}",
+                )
+            )
+            await _safe_upsert(session)
+        yield StreamError(
+            errorText=FRIENDLY_TRANSIENT_MSG,
+            code="transient_api_error",
+        )
+        return
+
+    # --- Magic keyword: fatal error (non-retryable) -------------------------
+    if _has_keyword(message, "__test_fatal_error__"):
+        yield StreamFinishStep()
+        error_msg = "Internal SDK error: model refused to respond"
+        # Persist non-retryable error marker
+        if session:
+            session.messages.append(
+                ChatMessage(
+                    role="assistant",
+                    content=f"{COPILOT_ERROR_PREFIX} {error_msg}",
+                )
+            )
+            await _safe_upsert(session)
+        yield StreamError(errorText=error_msg, code="sdk_error")
+        return
+
+    # --- Magic keyword: slow response ---------------------------------------
+    delay = 2.0 if _has_keyword(message, "__test_slow_response__") else 0.1
+
+    # --- Normal dummy response ----------------------------------------------
     dummy_response = "I counted: 1... 2... 3. All done!"
     words = dummy_response.split()
 
+    yield StreamTextStart(id=text_block_id)
     for i, word in enumerate(words):
         # Add space except for last word
         text = word if i == len(words) - 1 else f"{word} "
         yield StreamTextDelta(id=text_block_id, delta=text)
-        # Small delay to simulate real streaming
-        await asyncio.sleep(0.1)
+        await asyncio.sleep(delay)
+    yield StreamTextEnd(id=text_block_id)
+
+    # Persist the assistant reply so it survives page refresh
+    if session:
+        session.messages.append(ChatMessage(role="assistant", content=dummy_response))
+        await _safe_upsert(session)
+
+    yield StreamFinishStep()
+    yield StreamFinish()

autogpt_platform/backend/backend/copilot/sdk/response_adapter.py

@@ -20,6 +20,7 @@ from claude_agent_sdk import (
     UserMessage,
 )
 
+from backend.copilot.constants import FRIENDLY_TRANSIENT_MSG, is_transient_api_error
 from backend.copilot.response_model import (
     StreamBaseResponse,
     StreamError,
@@ -214,10 +215,12 @@ class SDKResponseAdapter:
             if sdk_message.subtype == "success":
                 responses.append(StreamFinish())
             elif sdk_message.subtype in ("error", "error_during_execution"):
-                error_msg = sdk_message.result or "Unknown error"
-                responses.append(
-                    StreamError(errorText=str(error_msg), code="sdk_error")
-                )
+                raw_error = str(sdk_message.result or "Unknown error")
+                if is_transient_api_error(raw_error):
+                    error_text, code = FRIENDLY_TRANSIENT_MSG, "transient_api_error"
+                else:
+                    error_text, code = raw_error, "sdk_error"
+                responses.append(StreamError(errorText=error_text, code=code))
                 responses.append(StreamFinish())
             else:
                 logger.warning(

autogpt_platform/backend/backend/copilot/sdk/service.py

@@ -37,7 +37,13 @@ from backend.util.prompt import compress_context
 from backend.util.settings import Settings
 
 from ..config import ChatConfig
-from ..constants import COPILOT_ERROR_PREFIX, COPILOT_SYSTEM_PREFIX
+from ..constants import (
+    COPILOT_ERROR_PREFIX,
+    COPILOT_RETRYABLE_ERROR_PREFIX,
+    COPILOT_SYSTEM_PREFIX,
+    FRIENDLY_TRANSIENT_MSG,
+    is_transient_api_error,
+)
 from ..model import (
     ChatMessage,
     ChatSession,
@@ -90,6 +96,28 @@ logger = logging.getLogger(__name__)
 config = ChatConfig()
 
 
+def _append_error_marker(
+    session: ChatSession | None,
+    display_msg: str,
+    *,
+    retryable: bool = False,
+) -> None:
+    """Append a copilot error marker to *session* so it persists across refresh.
+
+    Args:
+        session: The chat session to append to (no-op if ``None``).
+        display_msg: User-visible error text.
+        retryable: If ``True``, use the retryable prefix so the frontend
+            shows a "Try Again" button.
+    """
+    if session is None:
+        return
+    prefix = COPILOT_RETRYABLE_ERROR_PREFIX if retryable else COPILOT_ERROR_PREFIX
+    session.messages.append(
+        ChatMessage(role="assistant", content=f"{prefix} {display_msg}")
+    )
+
+
 def _setup_langfuse_otel() -> None:
     """Configure OTEL tracing for the Claude Agent SDK → Langfuse.
 
@@ -158,7 +186,12 @@ def _resolve_sdk_model() -> str | None:
 
     Uses ``config.claude_agent_model`` if set, otherwise derives from
     ``config.model`` by stripping the OpenRouter provider prefix (e.g.,
-    ``"anthropic/claude-opus-4.6"`` → ``"claude-opus-4.6"``).
+    ``"anthropic/claude-opus-4.6"`` → ``"claude-opus-4-6"``).
+
+    OpenRouter uses dot-separated versions (``claude-opus-4.6``) while the
+    direct Anthropic API uses hyphen-separated versions (``claude-opus-4-6``).
+    Normalisation is only applied when the SDK will actually talk to
+    Anthropic directly (not through OpenRouter).
 
     When ``use_claude_code_subscription`` is enabled and no explicit
     ``claude_agent_model`` is set, returns ``None`` so the CLI uses the
@@ -170,7 +203,12 @@ def _resolve_sdk_model() -> str | None:
         return None
     model = config.model
     if "/" in model:
-        return model.split("/", 1)[1]
+        model = model.split("/", 1)[1]
+    # OpenRouter uses dots in versions (claude-opus-4.6) but the direct
+    # Anthropic API requires hyphens (claude-opus-4-6).  Only normalise
+    # when NOT routing through OpenRouter.
+    if not config.openrouter_active:
+        model = model.replace(".", "-")
     return model
 
 
@@ -209,61 +247,52 @@ def _build_sdk_env(
     session_id: str | None = None,
     user_id: str | None = None,
 ) -> dict[str, str]:
-    """Build env vars for the SDK CLI process.
+    """Build env vars for the SDK CLI subprocess.
 
-    Routes API calls through OpenRouter (or a custom base_url) using
-    the same ``config.api_key`` / ``config.base_url`` as the non-SDK path.
-    This gives per-call token and cost tracking on the OpenRouter dashboard.
-
-    When *session_id* is provided, an ``x-session-id`` custom header is
-    injected via ``ANTHROPIC_CUSTOM_HEADERS`` so that OpenRouter Broadcast
-    forwards traces (including cost/usage) to Langfuse for the
-    ``/api/v1/messages`` endpoint.
-
-    Only overrides ``ANTHROPIC_API_KEY`` when a valid proxy URL and auth
-    token are both present — otherwise returns an empty dict so the SDK
-    falls back to its default credentials.
+    Three modes (checked in order):
+    1. **Subscription** — clears all keys; CLI uses ``claude login`` auth.
+    2. **Direct Anthropic** — returns ``{}``; subprocess inherits
+       ``ANTHROPIC_API_KEY`` from the parent environment.
+    3. **OpenRouter** (default) — overrides base URL and auth token to
+       route through the proxy, with Langfuse trace headers.
     """
-    env: dict[str, str] = {}
-
+    # --- Mode 1: Claude Code subscription auth ---
     if config.use_claude_code_subscription:
-        # Claude Code subscription: let the CLI use its own logged-in auth.
-        # Explicitly clear API key env vars so the subprocess doesn't pick
-        # them up from the parent process and bypass subscription auth.
         _validate_claude_code_subscription()
-        env["ANTHROPIC_API_KEY"] = ""
-        env["ANTHROPIC_AUTH_TOKEN"] = ""
-        env["ANTHROPIC_BASE_URL"] = ""
-    elif config.api_key and config.base_url:
-        # Strip /v1 suffix — SDK expects the base URL without a version path
-        base = config.base_url.rstrip("/")
-        if base.endswith("/v1"):
-            base = base[:-3]
-        if not base or not base.startswith("http"):
-            # Invalid base_url — don't override SDK defaults
-            return env
-        env["ANTHROPIC_BASE_URL"] = base
-        env["ANTHROPIC_AUTH_TOKEN"] = config.api_key
-        # Must be explicitly empty so the CLI uses AUTH_TOKEN instead
-        env["ANTHROPIC_API_KEY"] = ""
+        return {
+            "ANTHROPIC_API_KEY": "",
+            "ANTHROPIC_AUTH_TOKEN": "",
+            "ANTHROPIC_BASE_URL": "",
+        }
+
+    # --- Mode 2: Direct Anthropic (no proxy hop) ---
+    # ``openrouter_active`` checks the flag *and* credential presence.
+    if not config.openrouter_active:
+        return {}
+
+    # --- Mode 3: OpenRouter proxy ---
+    # Strip /v1 suffix — SDK expects the base URL without a version path.
+    base = (config.base_url or "").rstrip("/")
+    if base.endswith("/v1"):
+        base = base[:-3]
+    env: dict[str, str] = {
+        "ANTHROPIC_BASE_URL": base,
+        "ANTHROPIC_AUTH_TOKEN": config.api_key or "",
+        "ANTHROPIC_API_KEY": "",  # force CLI to use AUTH_TOKEN
+    }
 
     # Inject broadcast headers so OpenRouter forwards traces to Langfuse.
-    # The ``x-session-id`` header is *required* for the Anthropic-native
-    # ``/messages`` endpoint — without it broadcast silently drops the
-    # trace even when org-level Langfuse integration is configured.
-    def _safe(value: str) -> str:
-        """Strip CR/LF to prevent header injection, then truncate."""
-        return value.replace("\r", "").replace("\n", "").strip()[:128]
+    def _safe(v: str) -> str:
+        """Sanitise a header value: strip newlines/whitespace and cap length."""
+        return v.replace("\r", "").replace("\n", "").strip()[:128]
 
-    headers: list[str] = []
+    parts = []
     if session_id:
-        headers.append(f"x-session-id: {_safe(session_id)}")
+        parts.append(f"x-session-id: {_safe(session_id)}")
     if user_id:
-        headers.append(f"x-user-id: {_safe(user_id)}")
-    # Only inject headers when routing through OpenRouter/proxy — they're
-    # meaningless (and leak internal IDs) when using subscription mode.
-    if headers and env.get("ANTHROPIC_BASE_URL"):
-        env["ANTHROPIC_CUSTOM_HEADERS"] = "\n".join(headers)
+        parts.append(f"x-user-id: {_safe(user_id)}")
+    if parts:
+        env["ANTHROPIC_CUSTOM_HEADERS"] = "\n".join(parts)
 
     return env
 
@@ -655,13 +684,17 @@ async def stream_chat_completion_sdk(
     # Type narrowing: session is guaranteed ChatSession after the check above
     session = cast(ChatSession, session)
 
-    # Clean up stale error markers from previous turn before starting new turn
-    # If the last message contains an error marker, remove it (user is retrying)
-    if (
+    # Clean up ALL trailing error markers from previous turn before starting
+    # a new turn.  Multiple markers can accumulate when a mid-stream error is
+    # followed by a cleanup error in __aexit__ (both append a marker).
+    while (
         len(session.messages) > 0
         and session.messages[-1].role == "assistant"
         and session.messages[-1].content
-        and COPILOT_ERROR_PREFIX in session.messages[-1].content
+        and (
+            COPILOT_ERROR_PREFIX in session.messages[-1].content
+            or COPILOT_RETRYABLE_ERROR_PREFIX in session.messages[-1].content
+        )
     ):
         logger.info(
             "[SDK] [%s] Removing stale error marker from previous turn",
@@ -806,7 +839,7 @@ async def stream_chat_completion_sdk(
                 )
             except Exception as transcript_err:
                 logger.warning(
-                    "%s Transcript download failed, continuing without " "--resume: %s",
+                    "%s Transcript download failed, continuing without --resume: %s",
                     log_prefix,
                     transcript_err,
                 )
@@ -829,7 +862,7 @@ async def stream_chat_completion_sdk(
             is_valid = validate_transcript(dl.content)
             dl_lines = dl.content.strip().split("\n") if dl.content else []
             logger.info(
-                "%s Downloaded transcript: %dB, %d lines, " "msg_count=%d, valid=%s",
+                "%s Downloaded transcript: %dB, %d lines, msg_count=%d, valid=%s",
                 log_prefix,
                 len(dl.content),
                 len(dl_lines),
@@ -1048,23 +1081,36 @@ async def stream_chat_completion_sdk(
                         # Exception in receive_response() — capture it
                         # so the session can still be saved and the
                         # frontend gets a clean finish.
-                        logger.error(
+                        if is_transient_api_error(str(stream_err)):
+                            log, display, code = (
+                                logger.warning,
+                                FRIENDLY_TRANSIENT_MSG,
+                                "transient_api_error",
+                            )
+                        else:
+                            log, display, code = (
+                                logger.error,
+                                f"SDK stream error: {stream_err}",
+                                "sdk_stream_error",
+                            )
+
+                        log(
                             "%s Stream error from SDK: %s",
                             log_prefix,
                             stream_err,
                             exc_info=True,
                         )
                         ended_with_stream_error = True
-
-                        yield StreamError(
-                            errorText=f"SDK stream error: {stream_err}",
-                            code="sdk_stream_error",
+                        _append_error_marker(
+                            session,
+                            display,
+                            retryable=(code == "transient_api_error"),
                         )
+                        yield StreamError(errorText=display, code=code)
                         break
 
                     logger.info(
-                        "%s Received: %s %s "
-                        "(unresolved=%d, current=%d, resolved=%d)",
+                        "%s Received: %s %s (unresolved=%d, current=%d, resolved=%d)",
                         log_prefix,
                         type(sdk_msg).__name__,
                         getattr(sdk_msg, "subtype", ""),
@@ -1078,15 +1124,42 @@ async def stream_chat_completion_sdk(
                     # so we can debug Anthropic API 400s surfaced by the CLI.
                     sdk_error = getattr(sdk_msg, "error", None)
                     if isinstance(sdk_msg, AssistantMessage) and sdk_error:
+                        error_text = str(sdk_error)
+                        error_preview = str(sdk_msg.content)[:500]
                         logger.error(
                             "[SDK] [%s] AssistantMessage has error=%s, "
                             "content_blocks=%d, content_preview=%s",
                             session_id[:12],
                             sdk_error,
                             len(sdk_msg.content),
-                            str(sdk_msg.content)[:500],
+                            error_preview,
                         )
 
+                        # Intercept transient API errors (socket closed,
+                        # ECONNRESET) — replace the raw message with a
+                        # user-friendly error text and use the retryable
+                        # error prefix so the frontend shows a retry button.
+                        # Check both the error field and content for patterns.
+                        if is_transient_api_error(error_text) or is_transient_api_error(
+                            error_preview
+                        ):
+                            logger.warning(
+                                "%s Transient Anthropic API error detected, "
+                                "suppressing raw error text",
+                                log_prefix,
+                            )
+                            ended_with_stream_error = True
+                            _append_error_marker(
+                                session,
+                                FRIENDLY_TRANSIENT_MSG,
+                                retryable=True,
+                            )
+                            yield StreamError(
+                                errorText=FRIENDLY_TRANSIENT_MSG,
+                                code="transient_api_error",
+                            )
+                            break
+
                     # Race-condition fix: SDK hooks (PostToolUse) are
                     # executed asynchronously via start_soon() — the next
                     # message can arrive before the hook stashes output.
@@ -1212,7 +1285,7 @@ async def stream_chat_completion_sdk(
                                 extra,
                             )
 
-                        # Log errors being sent to frontend
+                        # Persist error markers so they survive page refresh
                         if isinstance(response, StreamError):
                             logger.error(
                                 "%s Sending error to frontend: %s (code=%s)",
@@ -1220,6 +1293,12 @@ async def stream_chat_completion_sdk(
                                 response.errorText,
                                 response.code,
                             )
+                            _append_error_marker(
+                                session,
+                                response.errorText,
+                                retryable=(response.code == "transient_api_error"),
+                            )
+                            ended_with_stream_error = True
 
                         yield response
 
@@ -1434,14 +1513,18 @@ async def stream_chat_completion_sdk(
             else:
                 logger.error("%s Error: %s", log_prefix, error_msg, exc_info=True)
 
-        # Append error marker to session (non-invasive text parsing approach)
-        # The finally block will persist the session with this error marker
-        if session:
-            session.messages.append(
-                ChatMessage(
-                    role="assistant", content=f"{COPILOT_ERROR_PREFIX} {error_msg}"
-                )
-            )
+        is_transient = is_transient_api_error(error_msg)
+        if is_transient:
+            display_msg, code = FRIENDLY_TRANSIENT_MSG, "transient_api_error"
+        else:
+            display_msg, code = error_msg, "sdk_error"
+
+        # Append error marker to session (non-invasive text parsing approach).
+        # The finally block will persist the session with this error marker.
+        # Skip if a marker was already appended inside the stream loop
+        # (ended_with_stream_error) to avoid duplicate stale markers.
+        if not ended_with_stream_error:
+            _append_error_marker(session, display_msg, retryable=is_transient)
             logger.debug(
                 "%s Appended error marker, will be persisted in finally",
                 log_prefix,
@@ -1453,10 +1536,7 @@ async def stream_chat_completion_sdk(
             isinstance(e, RuntimeError) and "cancel scope" in str(e)
         )
         if not is_cancellation:
-            yield StreamError(
-                errorText=error_msg,
-                code="sdk_error",
-            )
+            yield StreamError(errorText=display_msg, code=code)
 
         raise
     finally:

autogpt_platform/backend/backend/copilot/sdk/service_test.py

@@ -8,7 +8,7 @@ from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
 
-from .service import _prepare_file_attachments
+from .service import _prepare_file_attachments, _resolve_sdk_model
 
 
 @dataclass
@@ -288,3 +288,127 @@ class TestPromptSupplement:
             # Count how many times this tool appears as a bullet point
             count = docs.count(f"- **`{tool_name}`**")
             assert count == 1, f"Tool '{tool_name}' appears {count} times (should be 1)"
+
+
+# ---------------------------------------------------------------------------
+# Env vars that ChatConfig validators read — must be cleared so explicit
+# constructor values are used.
+# ---------------------------------------------------------------------------
+_CONFIG_ENV_VARS = (
+    "CHAT_USE_OPENROUTER",
+    "CHAT_API_KEY",
+    "OPEN_ROUTER_API_KEY",
+    "OPENAI_API_KEY",
+    "CHAT_BASE_URL",
+    "OPENROUTER_BASE_URL",
+    "OPENAI_BASE_URL",
+    "CHAT_USE_CLAUDE_CODE_SUBSCRIPTION",
+    "CHAT_USE_CLAUDE_AGENT_SDK",
+)
+
+
+@pytest.fixture()
+def _clean_config_env(monkeypatch: pytest.MonkeyPatch) -> None:
+    for var in _CONFIG_ENV_VARS:
+        monkeypatch.delenv(var, raising=False)
+
+
+class TestResolveSdkModel:
+    """Tests for _resolve_sdk_model — model ID resolution for the SDK CLI."""
+
+    def test_openrouter_active_keeps_dots(self, monkeypatch, _clean_config_env):
+        """When OpenRouter is fully active, model keeps dot-separated version."""
+        from backend.copilot import config as cfg_mod
+
+        cfg = cfg_mod.ChatConfig(
+            model="anthropic/claude-opus-4.6",
+            claude_agent_model=None,
+            use_openrouter=True,
+            api_key="or-key",
+            base_url="https://openrouter.ai/api/v1",
+            use_claude_code_subscription=False,
+        )
+        monkeypatch.setattr("backend.copilot.sdk.service.config", cfg)
+        assert _resolve_sdk_model() == "claude-opus-4.6"
+
+    def test_openrouter_disabled_normalizes_to_hyphens(
+        self, monkeypatch, _clean_config_env
+    ):
+        """When OpenRouter is disabled, dots are replaced with hyphens."""
+        from backend.copilot import config as cfg_mod
+
+        cfg = cfg_mod.ChatConfig(
+            model="anthropic/claude-opus-4.6",
+            claude_agent_model=None,
+            use_openrouter=False,
+            api_key=None,
+            base_url=None,
+            use_claude_code_subscription=False,
+        )
+        monkeypatch.setattr("backend.copilot.sdk.service.config", cfg)
+        assert _resolve_sdk_model() == "claude-opus-4-6"
+
+    def test_openrouter_enabled_but_missing_key_normalizes(
+        self, monkeypatch, _clean_config_env
+    ):
+        """When OpenRouter is enabled but api_key is missing, falls back to
+        direct Anthropic and normalizes dots to hyphens."""
+        from backend.copilot import config as cfg_mod
+
+        cfg = cfg_mod.ChatConfig(
+            model="anthropic/claude-opus-4.6",
+            claude_agent_model=None,
+            use_openrouter=True,
+            api_key=None,
+            base_url="https://openrouter.ai/api/v1",
+            use_claude_code_subscription=False,
+        )
+        monkeypatch.setattr("backend.copilot.sdk.service.config", cfg)
+        assert _resolve_sdk_model() == "claude-opus-4-6"
+
+    def test_explicit_claude_agent_model_takes_precedence(
+        self, monkeypatch, _clean_config_env
+    ):
+        """When claude_agent_model is explicitly set, it is returned as-is."""
+        from backend.copilot import config as cfg_mod
+
+        cfg = cfg_mod.ChatConfig(
+            model="anthropic/claude-opus-4.6",
+            claude_agent_model="claude-sonnet-4-5-20250514",
+            use_openrouter=True,
+            api_key="or-key",
+            base_url="https://openrouter.ai/api/v1",
+            use_claude_code_subscription=False,
+        )
+        monkeypatch.setattr("backend.copilot.sdk.service.config", cfg)
+        assert _resolve_sdk_model() == "claude-sonnet-4-5-20250514"
+
+    def test_subscription_mode_returns_none(self, monkeypatch, _clean_config_env):
+        """When using Claude Code subscription, returns None (CLI picks model)."""
+        from backend.copilot import config as cfg_mod
+
+        cfg = cfg_mod.ChatConfig(
+            model="anthropic/claude-opus-4.6",
+            claude_agent_model=None,
+            use_openrouter=False,
+            api_key=None,
+            base_url=None,
+            use_claude_code_subscription=True,
+        )
+        monkeypatch.setattr("backend.copilot.sdk.service.config", cfg)
+        assert _resolve_sdk_model() is None
+
+    def test_model_without_provider_prefix(self, monkeypatch, _clean_config_env):
+        """When model has no provider prefix, it still normalizes correctly."""
+        from backend.copilot import config as cfg_mod
+
+        cfg = cfg_mod.ChatConfig(
+            model="claude-opus-4.6",
+            claude_agent_model=None,
+            use_openrouter=False,
+            api_key=None,
+            base_url=None,
+            use_claude_code_subscription=False,
+        )
+        monkeypatch.setattr("backend.copilot.sdk.service.config", cfg)
+        assert _resolve_sdk_model() == "claude-opus-4-6"

autogpt_platform/backend/backend/copilot/test_copilot_e2e.py

@@ -4,11 +4,12 @@ These tests verify the complete copilot flow using dummy implementations
 for agent generator and SDK service, allowing automated testing without
 external LLM calls.
 
-Enable test mode with COPILOT_TEST_MODE=true environment variable.
+Enable test mode with CHAT_TEST_MODE=true environment variable (or in .env).
 
-Note: StreamFinish is NOT emitted by the dummy service — it is published
-by mark_session_completed in the processor layer.  These tests only cover
-the service-level streaming output (StreamStart + StreamTextDelta).
+The dummy service emits the full AI SDK protocol event sequence:
+StreamStart → StreamStartStep → StreamTextStart → StreamTextDelta(s) →
+StreamTextEnd → StreamFinishStep → StreamFinish.
+The processor skips StreamFinish and publishes its own via mark_session_completed.
 """
 
 import asyncio
@@ -20,9 +21,14 @@ import pytest
 from backend.copilot.model import ChatMessage, ChatSession, upsert_chat_session
 from backend.copilot.response_model import (
     StreamError,
+    StreamFinish,
+    StreamFinishStep,
     StreamHeartbeat,
     StreamStart,
+    StreamStartStep,
     StreamTextDelta,
+    StreamTextEnd,
+    StreamTextStart,
 )
 from backend.copilot.sdk.dummy import stream_chat_completion_dummy
 
@@ -30,9 +36,9 @@ from backend.copilot.sdk.dummy import stream_chat_completion_dummy
 @pytest.fixture(autouse=True)
 def enable_test_mode():
     """Enable test mode for all tests in this module."""
-    os.environ["COPILOT_TEST_MODE"] = "true"
+    os.environ["CHAT_TEST_MODE"] = "true"
     yield
-    os.environ.pop("COPILOT_TEST_MODE", None)
+    os.environ.pop("CHAT_TEST_MODE", None)
 
 
 @pytest.mark.asyncio
@@ -110,9 +116,14 @@ async def test_streaming_event_types():
     ):
         event_types.add(type(event).__name__)
 
-    # Required event types (StreamFinish is published by processor, not service)
+    # Required event types for full AI SDK protocol
     assert "StreamStart" in event_types, "Missing StreamStart"
+    assert "StreamStartStep" in event_types, "Missing StreamStartStep"
+    assert "StreamTextStart" in event_types, "Missing StreamTextStart"
     assert "StreamTextDelta" in event_types, "Missing StreamTextDelta"
+    assert "StreamTextEnd" in event_types, "Missing StreamTextEnd"
+    assert "StreamFinishStep" in event_types, "Missing StreamFinishStep"
+    assert "StreamFinish" in event_types, "Missing StreamFinish"
 
     print(f"✅ Event types: {sorted(event_types)}")
 
@@ -175,16 +186,17 @@ async def test_streaming_heartbeat_timing():
 
 @pytest.mark.asyncio
 async def test_error_handling():
-    """Test that errors are properly formatted and sent."""
-    # This would require a dummy that can trigger errors
-    # For now, just verify error event structure
-
+    """Test that error events have correct SSE structure."""
     error = StreamError(errorText="Test error", code="test_error")
     assert error.errorText == "Test error"
     assert error.code == "test_error"
-    assert str(error.type.value) in ["error", "error"]
 
-    print("✅ Error structure verified")
+    # Verify to_sse() strips code (AI SDK protocol compliance)
+    sse = error.to_sse()
+    assert '"errorText"' in sse
+    assert '"code"' not in sse, "to_sse() must strip code field for AI SDK"
+
+    print("✅ Error structure verified (code stripped in SSE)")
 
 
 @pytest.mark.asyncio
@@ -326,20 +338,85 @@ async def test_stream_completeness():
     ):
         events.append(event)
 
-    # Check for required events (StreamFinish is published by processor)
-    has_start = any(isinstance(e, StreamStart) for e in events)
-    has_text = any(isinstance(e, StreamTextDelta) for e in events)
-
-    assert has_start, "Stream must include StreamStart"
-    assert has_text, "Stream must include text deltas"
+    # Check for all required event types
+    assert any(isinstance(e, StreamStart) for e in events), "Missing StreamStart"
+    assert any(
+        isinstance(e, StreamStartStep) for e in events
+    ), "Missing StreamStartStep"
+    assert any(
+        isinstance(e, StreamTextStart) for e in events
+    ), "Missing StreamTextStart"
+    assert any(
+        isinstance(e, StreamTextDelta) for e in events
+    ), "Missing StreamTextDelta"
+    assert any(isinstance(e, StreamTextEnd) for e in events), "Missing StreamTextEnd"
+    assert any(
+        isinstance(e, StreamFinishStep) for e in events
+    ), "Missing StreamFinishStep"
+    assert any(isinstance(e, StreamFinish) for e in events), "Missing StreamFinish"
 
     # Verify exactly one start
     start_count = sum(1 for e in events if isinstance(e, StreamStart))
     assert start_count == 1, f"Should have exactly 1 StreamStart, got {start_count}"
 
-    print(
-        f"✅ Completeness: 1 start, {sum(1 for e in events if isinstance(e, StreamTextDelta))} text deltas"
-    )
+    print(f"✅ Completeness: {len(events)} events, full protocol sequence")
+
+
+@pytest.mark.asyncio
+async def test_transient_error_shows_retryable():
+    """Test __test_transient_error__ yields partial text then retryable StreamError."""
+    events = []
+
+    async for event in stream_chat_completion_dummy(
+        session_id="test-transient",
+        message="please fail __test_transient_error__",
+        is_user_message=True,
+        user_id="test-user",
+    ):
+        events.append(event)
+
+    # Should start with StreamStart
+    assert isinstance(events[0], StreamStart)
+
+    # Should have some partial text before the error
+    text_events = [e for e in events if isinstance(e, StreamTextDelta)]
+    assert len(text_events) > 0, "Should stream partial text before error"
+
+    # Should end with StreamError
+    error_events = [e for e in events if isinstance(e, StreamError)]
+    assert len(error_events) == 1, "Should have exactly one StreamError"
+    assert error_events[0].code == "transient_api_error"
+    assert "connection interrupted" in error_events[0].errorText.lower()
+
+    print(f"✅ Transient error: {len(text_events)} partial deltas + retryable error")
+
+
+@pytest.mark.asyncio
+async def test_fatal_error_not_retryable():
+    """Test __test_fatal_error__ yields StreamError without retryable code."""
+    events = []
+
+    async for event in stream_chat_completion_dummy(
+        session_id="test-fatal",
+        message="__test_fatal_error__",
+        is_user_message=True,
+        user_id="test-user",
+    ):
+        events.append(event)
+
+    assert isinstance(events[0], StreamStart)
+
+    # Should have StreamError with sdk_error code (not transient)
+    error_events = [e for e in events if isinstance(e, StreamError)]
+    assert len(error_events) == 1
+    assert error_events[0].code == "sdk_error"
+    assert "transient" not in error_events[0].code
+
+    # Should NOT have any text deltas (fatal errors fail immediately)
+    text_events = [e for e in events if isinstance(e, StreamTextDelta)]
+    assert len(text_events) == 0, "Fatal error should not stream any text"
+
+    print("✅ Fatal error: immediate error, no partial text")
 
 
 @pytest.mark.asyncio
@@ -395,6 +472,8 @@ if __name__ == "__main__":
     asyncio.run(test_message_deduplication())
     asyncio.run(test_event_ordering())
     asyncio.run(test_stream_completeness())
+    asyncio.run(test_transient_error_shows_retryable())
+    asyncio.run(test_fatal_error_not_retryable())
     asyncio.run(test_text_delta_consistency())
 
     print("=" * 60)

autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatContainer/ChatContainer.tsx

@@ -2,6 +2,7 @@
 import { ChatInput } from "@/app/(platform)/copilot/components/ChatInput/ChatInput";
 import { UIDataTypes, UIMessage, UITools } from "ai";
 import { LayoutGroup, motion } from "framer-motion";
+import { useCallback } from "react";
 import { ChatMessagesContainer } from "../ChatMessagesContainer/ChatMessagesContainer";
 import { CopilotChatActionsProvider } from "../CopilotChatActionsProvider/CopilotChatActionsProvider";
 import { EmptySession } from "../EmptySession/EmptySession";
@@ -49,6 +50,20 @@ export const ChatContainer = ({
     !!isSessionError;
   const inputLayoutId = "copilot-2-chat-input";
 
+  // Retry: re-send the last user message (used by ErrorCard on transient errors)
+  const handleRetry = useCallback(() => {
+    const lastUserMsg = [...messages].reverse().find((m) => m.role === "user");
+    const lastText = lastUserMsg?.parts
+      .filter(
+        (p): p is Extract<typeof p, { type: "text" }> => p.type === "text",
+      )
+      .map((p) => p.text)
+      .join("");
+    if (lastText) {
+      onSend(lastText);
+    }
+  }, [messages, onSend]);
+
   return (
     <CopilotChatActionsProvider onSend={onSend}>
       <LayoutGroup id="copilot-2-chat-layout">
@@ -61,6 +76,7 @@ export const ChatContainer = ({
                 error={error}
                 isLoading={isLoadingSession}
                 sessionID={sessionId}
+                onRetry={handleRetry}
               />
               <motion.div
                 initial={{ opacity: 0 }}

autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatMessagesContainer/ChatMessagesContainer.tsx

@@ -31,11 +31,13 @@ interface Props {
   error: Error | undefined;
   isLoading: boolean;
   sessionID?: string | null;
+  onRetry?: () => void;
 }
 
 function renderSegments(
   segments: RenderSegment[],
   messageID: string,
+  onRetry?: () => void,
 ): React.ReactNode[] {
   return segments.map((seg, segIdx) => {
     if (seg.kind === "collapsed-group") {
@@ -47,6 +49,7 @@ function renderSegments(
         part={seg.part}
         messageID={messageID}
         partIndex={seg.index}
+        onRetry={onRetry}
       />
     );
   });
@@ -102,6 +105,7 @@ export function ChatMessagesContainer({
   error,
   isLoading,
   sessionID,
+  onRetry,
 }: Props) {
   const lastMessage = messages[messages.length - 1];
   const graphExecId = useMemo(() => extractGraphExecId(messages), [messages]);
@@ -161,9 +165,12 @@ export function ChatMessagesContainer({
             (p): p is Extract<typeof p, { type: "text" }> => p.type === "text",
           );
           const lastTextPart = textParts[textParts.length - 1];
+          const markerType =
+            lastTextPart !== undefined
+              ? parseSpecialMarkers(lastTextPart.text).markerType
+              : null;
           const hasErrorMarker =
-            lastTextPart !== undefined &&
-            parseSpecialMarkers(lastTextPart.text).markerType === "error";
+            markerType === "error" || markerType === "retryable_error";
           const showActions =
             isLastInTurn &&
             !isCurrentlyStreaming &&
@@ -209,13 +216,18 @@ export function ChatMessagesContainer({
                   </ReasoningCollapse>
                 )}
                 {responseSegments
-                  ? renderSegments(responseSegments, message.id)
+                  ? renderSegments(
+                      responseSegments,
+                      message.id,
+                      isLastAssistant ? onRetry : undefined,
+                    )
                   : message.parts.map((part, i) => (
                       <MessagePartRenderer
                         key={`${message.id}-${i}`}
                         part={part}
                         messageID={message.id}
                         partIndex={i}
+                        onRetry={isLastAssistant ? onRetry : undefined}
                       />
                     ))}
                 {isLastInTurn && !isCurrentlyStreaming && (

autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatMessagesContainer/components/MessagePartRenderer.tsx

@@ -69,9 +69,15 @@ interface Props {
   part: UIMessage<unknown, UIDataTypes, UITools>["parts"][number];
   messageID: string;
   partIndex: number;
+  onRetry?: () => void;
 }
 
-export function MessagePartRenderer({ part, messageID, partIndex }: Props) {
+export function MessagePartRenderer({
+  part,
+  messageID,
+  partIndex,
+  onRetry,
+}: Props) {
   const key = `${messageID}-${partIndex}`;
 
   switch (part.type) {
@@ -80,7 +86,7 @@ export function MessagePartRenderer({ part, messageID, partIndex }: Props) {
         part.text,
       );
 
-      if (markerType === "error") {
+      if (markerType === "error" || markerType === "retryable_error") {
         const lowerMarker = markerText.toLowerCase();
         const isCancellation =
           lowerMarker === "operation cancelled" ||
@@ -100,6 +106,7 @@ export function MessagePartRenderer({ part, messageID, partIndex }: Props) {
             key={key}
             responseError={{ message: markerText }}
             context="execution"
+            onRetry={markerType === "retryable_error" ? onRetry : undefined}
           />
         );
       }

autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatMessagesContainer/helpers.ts

@@ -172,16 +172,22 @@ export function getTurnMessages(
 // The hex suffix makes it virtually impossible for an LLM to accidentally
 // produce these strings in normal conversation.
 const COPILOT_ERROR_PREFIX = "[__COPILOT_ERROR_f7a1__]";
+const COPILOT_RETRYABLE_ERROR_PREFIX = "[__COPILOT_RETRYABLE_ERROR_a9c2__]";
 const COPILOT_SYSTEM_PREFIX = "[__COPILOT_SYSTEM_e3b0__]";
 
-export type MarkerType = "error" | "system" | null;
+export type MarkerType = "error" | "retryable_error" | "system" | null;
 
 /** Escape all regex special characters in a string. */
 function escapeRegExp(s: string): string {
   return s.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
 }
 
-// Pre-compiled marker regexes (avoids re-creating on every call / render)
+// Pre-compiled marker regexes (avoids re-creating on every call / render).
+// Retryable check must come first since it's more specific.
+const RETRYABLE_ERROR_MARKER_RE = new RegExp(
+  `${escapeRegExp(COPILOT_RETRYABLE_ERROR_PREFIX)}\\s*(.+?)$`,
+  "s",
+);
 const ERROR_MARKER_RE = new RegExp(
   `${escapeRegExp(COPILOT_ERROR_PREFIX)}\\s*(.+?)$`,
   "s",
@@ -196,6 +202,15 @@ export function parseSpecialMarkers(text: string): {
   markerText: string;
   cleanText: string;
 } {
+  const retryableMatch = text.match(RETRYABLE_ERROR_MARKER_RE);
+  if (retryableMatch) {
+    return {
+      markerType: "retryable_error",
+      markerText: retryableMatch[1].trim(),
+      cleanText: text.replace(retryableMatch[0], "").trim(),
+    };
+  }
+
   const errorMatch = text.match(ERROR_MARKER_RE);
   if (errorMatch) {
     return {

autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotStream.ts

@@ -222,12 +222,17 @@ export function useCopilotStream({
         return;
       }
 
-      // Only reconnect on network errors (not HTTP errors), and never
-      // reconnect when the user explicitly stopped the stream.
+      // Reconnect on network errors or transient API errors so the
+      // persisted retryable-error marker is loaded and the "Try Again"
+      // button appears.  Without this, transient errors only show in the
+      // onError callback (where StreamError strips the retryable prefix).
       if (isUserStoppingRef.current) return;
       const isNetworkError =
         error.name === "TypeError" || error.name === "AbortError";
-      if (isNetworkError) {
+      const isTransientApiError = errorDetail.includes(
+        "connection interrupted",
+      );
+      if (isNetworkError || isTransientApiError) {
         handleReconnect(sessionId);
       }
     },