NousResearch/hermes-agent

Gemini reasoning config mapping and visibility bugs

Cluster 16941 · agent. Data as of 2026-10-23 11:54 UTC

3 open items. Fix merged in #16999. Check it covers your case.

open issue

open PRs

items grouped

Known?: Yes. 6 related items grouped here, 1 issue already closed.
Fix event: A fix was merged to main in PR #16999. Check it covers your case. No release data is tracked.
Tracker: The maintainers' tracker marks this partially resolved, deciding PR #16999. That is separate from the fix event.
Still open: 1 issue and 2 PRs.

Next step: read #16999, then check whether your report is one of the 1 still open.

Read #16999 Pick up a Lab task

These issues and fixes address how reasoning/thinking configuration is bridged and mapped for Gemini and Vertex routes, including thought summaries incorrectly appearing in user-visible content and reasoning effort being incorrectly coupled with thought-summary visibility.

Open issues1

Issue#64035[Bug]: Gemini/Vertex thought summaries leak into user-visible content with display.show_reasoning: false (effort and visibility coupled via include_thoughts)

Open PRs2

PR#64036fix(gemini): decouple reasoning effort from thought-summary visibility
PR#91008fix(vertex): route Gemini thought deltas to the reasoning stream, not visible content

Fixes on record

Merged to main. A stamp counts the issues a fix is linked to.

#16999 fix(gemini): bridge reasoning_config into thinking_config for chat-completions routes

merged to main

files: agent/transports/chat_completions.py, run_agent.py, tests/agent/transports/test_chat_completions.py

Show the diff

agent/transports/chat_completions.py

@@ -18,6 +18,58 @@
 from agent.transports.types import NormalizedResponse, ToolCall, Usage
 
 
+def _build_gemini_thinking_config(model: str, reasoning_config: dict | None) -> dict | None:
+    """Translate Hermes/OpenRouter-style reasoning config to Gemini thinkingConfig.
+
+    Gemini native/cloud-code adapters do not read ``extra_body.reasoning``.
+    They only inspect ``extra_body.thinking_config`` / ``thinkingConfig`` and
+    then request thought parts with ``includeThoughts`` enabled.
+    """
+    if reasoning_config is None or not isinstance(reasoning_config, dict):
+        return None
+
+    if reasoning_config.get("enabled") is False:
+        # Gemini can hide thought parts even when internal thinking still
+        # happens; omit thinkingLevel to avoid model-specific validation quirks.
+        return {"includeThoughts": False}
+
+    effort = str(reasoning_config.get("effort", "medium") or "medium").strip().lower()
+    if effort == "none":
+        return {"includeThoughts": False}
+
+    thinking_config: Dict[str, Any] = {"includeThoughts": True}
+    normalized_model = (model or "").strip().lower()
+    if normalized_model.startswith("google/"):
+        normalized_model = normalized_model.split("/", 1)[1]
+
+    # Gemini 2.5 accepts thinkingBudget; don't guess a budget from Hermes'
+    # coarse effort levels. ``includeThoughts`` alone is enough to surface
+    # thought parts without risking request validation errors.
+    if normalized_model.startswith("gemini-2.5-"):
+        return thinking_config
+
+    if effort not in {"minimal", "low", "medium", "high", "xhigh"}:
+        effort = "medium"
+
+    # Gemini 3 Flash documents low/medium/high thinking levels; Gemini 3 Pro
+    # is stricter (low/high). Clamp Hermes' wider effort set to what each
+    # family accepts so we never forward an undocumented level verbatim.
+    if normalized_model.startswith(("gemini-3", "gemini-3.1")):
+        if "flash" in normalized_model:
+            if effort in {"minimal", "low"}:
+                thinking_config["thinkingLevel"] = "low"
+            elif effort in {"high", "xhigh"}:
+                thinking_config["thinkingLevel"] = "high"
+            else:
+                thinking_config["thinkingLevel"] = "medium"
+        elif "pro" in normalized_model:
+            thinking_config["thinkingLevel"] = (
+                "high" if effort in {"high", "xhigh"} else "low"
+            )
+
+    return thinking_config
+
+
 class ChatCompletionsTransport(ProviderTransport):
     """Transport for api_mode='chat_completions'.
 
@@ -241,6 +293,7 @@ def build_kwargs(
         is_openrouter = params.get("is_openrouter", False)
         is_nous = params.get("is_nous", False)
         is_github_models = params.get("is_github_models", False)
+        provider_name = str(params.get("provider_name") or "").strip().lower()
 
         provider_prefs = params.get("provider_preferences")
         if provider_prefs and is_openrouter:
@@ -293,6 +346,11 @@ def build_kwargs(
         if is_qwen:
             extra_body["vl_high_resolution_images"] = True
 
+        if provider_name in {"gemini", "google-gemini-cli"}:
+            thinking_config = _build_gemini_thinking_config(model, reasoning_config)
+            if thinking_config:
+                extra_body["thinking_config"] = thinking_config
+
         # Merge any pre-built extra_body additions
         additions = params.get("extra_body_additions")
         if additions:

run_agent.py

@@ -8094,6 +8094,7 @@ def _build_api_kwargs(self, api_messages: list) -> dict:
             supports_reasoning=self._supports_reasoning_extra_body(),
             github_reasoning_extra=self._github_models_reasoning_extra_body() if _is_gh else None,
             anthropic_max_output=_ant_max,
+            provider_name=self.provider,
         )
 
     def _supports_reasoning_extra_body(self) -> bool:

tests/agent/transports/test_chat_completions.py

@@ -4,7 +4,7 @@
 from types import SimpleNamespace
 
 from agent.transports import get_transport
-from agent.transports.types import NormalizedResponse, ToolCall
+from agent.transports.types import NormalizedResponse
 
 
 @pytest.fixture
@@ -122,6 +122,90 @@ def test_custom_think_false(self, transport):
         )
         assert kw["extra_body"]["think"] is False
 
+    def test_gemini_without_explicit_reasoning_config_keeps_existing_behavior(self, transport):
+        msgs = [{"role": "user", "content": "Hi"}]
+        kw = transport.build_kwargs(
+            model="gemini-3-flash-preview",
+            messages=msgs,
+            provider_name="gemini",
+        )
+        assert "thinking_config" not in kw.get("extra_body", {})
+
+    def test_gemini_flash_reasoning_maps_to_thinking_config(self, transport):
+        msgs = [{"role": "user", "content": "Hi"}]
+        kw = transport.build_kwargs(
+            model="gemini-3-flash-preview",
+            messages=msgs,
+            provider_name="gemini",
+            reasoning_config={"enabled": True, "effort": "high"},
+        )
+        assert kw["extra_body"]["thinking_config"] == {
+            "includeThoughts": True,
+            "thinkingLevel": "high",
+        }
+
+    def test_gemini_25_reasoning_only_enables_visible_thoughts(self, transport):
+        msgs = [{"role": "user", "content": "Hi"}]
+        kw = transport.build_kwargs(
+            model="gemini-2.5-flash",
+            messages=msgs,
+            provider_name="gemini",
+            reasoning_config={"enabled": True, "effort": "high"},
+        )
+        assert kw["extra_body"]["thinking_config"] == {
+            "includeThoughts": True,
+        }
+
+    def test_gemini_pro_reasoning_clamps_to_supported_levels(self, transport):
+        msgs = [{"role": "user", "content": "Hi"}]
+        kw = transport.build_kwargs(
+            model="google/gemini-3.1-pro-preview",
+            messages=msgs,
+            provider_name="gemini",
+            reasoning_config={"enabled": True, "effort": "medium"},
+        )
+        assert kw["extra_body"]["thinking_config"] == {
+            "includeThoughts": True,
+            "thinkingLevel": "low",
+        }
+
+    def test_gemini_disabled_reasoning_hides_thoughts(self, transport):
+        msgs = [{"role": "user", "content": "Hi"}]
+        kw = transport.build_kwargs(
+            model="gemini-3-flash-preview",
+            messages=msgs,
+            provider_name="gemini",
+            reasoning_config={"enabled": False},
+        )
+        assert kw["extra_body"]["thinking_config"] == {
+            "includeThoughts": False,
+        }
+
+    def test_gemini_xhigh_clamps_to_high(self, transport):
+        msgs = [{"role": "user", "content": "Hi"}]
+        kw = transport.build_kwargs(
+            model="gemini-3-flash-preview",
+            messages=msgs,
+            provider_name="gemini",
+            reasoning_config={"enabled": True, "effort": "xhigh"},
+        )
+        assert kw["extra_body"]["thinking_config"]["thinkingLevel"] == "high"
+
+    def test_gemini_flash_minimal_clamps_to_low(self, transport):
+        # Gemini 3 Flash documents low/medium/high; "minimal" isn't accepted,
+        # so clamp it down to "low" rather than forwarding it verbatim.
+        msgs = [{"role": "user", "content": "Hi"}]
+        kw = transport.build_kwargs(
+            model="gemini-3-flash-preview",
+            messages=msgs,
+            provider_name="gemini",
+            reasoning_config={"enabled": True, "effort": "minimal"},
+        )
+        assert kw["extra_body"]["thinking_config"] == {
+            "includeThoughts": True,
+            "thinkingLevel": "low",
+        }
+
     def test_max_tokens_with_fn(self, transport):
         msgs = [{"role": "user", "content": "Hi"}]
         kw = transport.build_kwargs(

#17605 fix(gemini): nest OpenAI-compat thinking config under google (salvage #17389)

merged to main

files: agent/auxiliary_client.py, agent/transports/chat_completions.py, run_agent.py, scripts/release.py, tests/agent/transports/test_chat_completions.py, tests/run_agent/test_provider_parity.py

Show the diff

agent/auxiliary_client.py

@@ -101,6 +101,14 @@ def __repr__(self):
 logger = logging.getLogger(__name__)
 
 
+def _safe_isinstance(obj: Any, maybe_type: Any) -> bool:
+    """Return False instead of raising when a patched symbol is not a type."""
+    try:
+        return isinstance(obj, maybe_type)
+    except TypeError:
+        return False
+
+
 def _extract_url_query_params(url: str):
     """Extract query params from URL, return (clean_url, default_query dict or None)."""
     parsed = urlparse(url)
@@ -861,20 +869,20 @@ def _maybe_wrap_anthropic(
     - The ``anthropic`` SDK is not installed (falls back to OpenAI wire).
     """
     # Already wrapped — don't double-wrap.
-    if isinstance(client_obj, AnthropicAuxiliaryClient):
+    if _safe_isinstance(client_obj, AnthropicAuxiliaryClient):
         return client_obj
     # Other specialized adapters we should never re-dispatch.
-    if isinstance(client_obj, CodexAuxiliaryClient):
+    if _safe_isinstance(client_obj, CodexAuxiliaryClient):
         return client_obj
     try:
         from agent.gemini_native_adapter import GeminiNativeClient
-        if isinstance(client_obj, GeminiNativeClient):
+        if _safe_isinstance(client_obj, GeminiNativeClient):
             return client_obj
     except ImportError:
         pass
     try:
         from agent.copilot_acp_client import CopilotACPClient
-        if isinstance(client_obj, CopilotACPClient):
+        if _safe_isinstance(client_obj, CopilotACPClient):
             return client_obj
     except ImportError:
         pass

agent/transports/chat_completions.py

@@ -20,12 +20,7 @@
 
 
 def _build_gemini_thinking_config(model: str, reasoning_config: dict | None) -> dict | None:
-    """Translate Hermes/OpenRouter-style reasoning config to Gemini thinkingConfig.
-
-    Gemini native/cloud-code adapters do not read ``extra_body.reasoning``.
-    They only inspect ``extra_body.thinking_config`` / ``thinkingConfig`` and
-    then request thought parts with ``includeThoughts`` enabled.
-    """
+    """Translate Hermes/OpenRouter-style reasoning config to Gemini thinkingConfig."""
     if reasoning_config is None or not isinstance(reasoning_config, dict):
         return None
 
@@ -71,6 +66,30 @@ def _build_gemini_thinking_config(model: str, reasoning_config: dict | None) ->
     return thinking_config
 
 
+def _snake_case_gemini_thinking_config(config: dict | None) -> dict | None:
+    """Convert Gemini thinking config keys to the OpenAI-compat field names."""
+    if not isinstance(config, dict) or not config:
+        return None
+
+    translated: Dict[str, Any] = {}
+    if isinstance(config.get("includeThoughts"), bool):
+        translated["include_thoughts"] = config["includeThoughts"]
+    if isinstance(config.get("thinkingLevel"), str) and config["thinkingLevel"].strip():
+        translated["thinking_level"] = config["thinkingLevel"].strip().lower()
+    if isinstance(config.get("thinkingBudget"), (int, float)):
+        translated["thinking_budget"] = int(config["thinkingBudget"])
+    return translated or None
+
+
+def _is_gemini_openai_compat_base_url(base_url: Any) -> bool:
+    normalized = str(base_url or "").strip().rstrip("/").lower()
+    if not normalized:
+        return False
+    if "generativelanguage.googleapis.com" not in normalized:
+        return False
+    return normalized.endswith("/openai")
+
+
 class ChatCompletionsTransport(ProviderTransport):
     """Transport for api_mode='chat_completions'.
 
@@ -309,6 +328,7 @@ def build_kwargs(
         is_nous = params.get("is_nous", False)
         is_github_models = params.get("is_github_models", False)
         provider_name = str(params.get("provider_name") or "").strip().lower()
+        base_url = params.get("base_url")
 
         provider_prefs = params.get("provider_preferences")
         if provider_prefs and is_openrouter:
@@ -362,7 +382,19 @@ def build_kwargs(
         if is_qwen:
             extra_body["vl_high_resolution_images"] = True
 
-        if provider_name in {"gemini", "google-gemini-cli"}:
+        if provider_name == "gemini":
+            raw_thinking_config = _build_gemini_thinking_config(model, reasoning_config)
+            if _is_gemini_openai_compat_base_url(base_url):
+                thinking_config = _snake_case_gemini_thinking_config(raw_thinking_config)
+                if thinking_config:
+                    openai_compat_extra = extra_body.get("extra_body", {})
+                    google_extra = openai_compat_extra.get("google", {})
+                    google_extra["thinking_config"] = thinking_config
+                    openai_compat_extra["google"] = google_extra
+                    extra_body["extra_body"] = openai_compat_extra
+            elif raw_thinking_config:
+                extra_body["thinking_config"] = raw_thinking_config
+        elif provider_name == "google-gemini-cli":
             thinking_config = _build_gemini_thinking_config(model, reasoning_config)
             if thinking_config:
                 extra_body["thinking_config"] = thinking_config

run_agent.py

@@ -8266,6 +8266,7 @@ def _build_api_kwargs(self, api_messages: list) -> dict:
             model=self.model,
             messages=_msgs_for_chat,
             tools=self.tools,
+            base_url=self.base_url,
             timeout=self._resolved_api_call_timeout(),
             max_tokens=[redacted:secret-assignment]
             ephemeral_max_output_tokens=[redacted:secret-assignment]

scripts/release.py

@@ -66,6 +66,7 @@
     "[redacted:email]": "dlkakbs",
     "[redacted:email]": "BeliefanX",
     "[redacted:email]": "Mind-Dragon",
+    "[redacted:email]": "Nanako0129",
     "[redacted:email]": "westers",
     "[redacted:email]": "yeyitech",
     "[redacted:email]": "beenherebefore",

tests/agent/transports/test_chat_completions.py

@@ -122,21 +122,25 @@ def test_custom_think_false(self, transport):
         )
         assert kw["extra_body"]["think"] is False
 
-    def test_gemini_without_explicit_reasoning_config_keeps_existing_behavior(self, transport):
+    def test_gemini_native_without_explicit_reasoning_config_keeps_existing_behavior(self, transport):
         msgs = [{"role": "user", "content": "Hi"}]
         kw = transport.build_kwargs(
             model="gemini-3-flash-preview",
             messages=msgs,
             provider_name="gemini",
+            base_url="https://generativelanguage.googleapis.com/v1beta",
         )
         assert "thinking_config" not in kw.get("extra_body", {})
+        assert "google" not in kw.get("extra_body", {})
+        assert "extra_body" not in kw.get("extra_body", {})
 
-    def test_gemini_flash_reasoning_maps_to_thinking_config(self, transport):
+    def test_gemini_native_flash_reasoning_maps_to_top_level_thinking_config(self, transport):
         msgs = [{"role": "user", "content": "Hi"}]
         kw = transport.build_kwargs(
             model="gemini-3-flash-preview",
             messages=msgs,
             provider_name="gemini",
+            base_url="https://generativelanguage.googleapis.com/v1beta",
             reasoning_config={"enabled": True, "effort": "high"},
         )
         assert kw["extra_body"]["thinking_config"] == {
@@ -144,52 +148,85 @@ def test_gemini_flash_reasoning_maps_to_thinking_config(self, transport):
             "thinkingLevel": "high",
         }
 
-    def test_gemini_25_reasoning_only_enables_visible_thoughts(self, transport):
+    def test_gemini_openai_compat_flash_reasoning_maps_to_nested_google_thinking_config(self, transport):
+        msgs = [{"role": "user", "content": "Hi"}]
+        kw = transport.build_kwargs(
+            model="gemini-3-flash-preview",
+            messages=msgs,
+            provider_name="gemini",
+            base_url="https://generativelanguage.googleapis.com/v1beta/openai",
+            reasoning_config={"enabled": True, "effort": "high"},
+        )
+        assert "thinking_config" not in kw["extra_body"]
+        assert kw["extra_body"]["extra_body"]["google"]["thinking_config"] == {
+            "include_thoughts": True,
+            "thinking_level": "high",
+        }
+
+    def test_gemini_native_25_reasoning_only_enables_visible_thoughts(self, transport):
         msgs = [{"role": "user", "content": "Hi"}]
         kw = transport.build_kwargs(
             model="gemini-2.5-flash",
             messages=msgs,
             provider_name="gemini",
+            base_url="https://generativelanguage.googleapis.com/v1beta",
             reasoning_config={"enabled": True, "effort": "high"},
         )
         assert kw["extra_body"]["thinking_config"] == {
             "includeThoughts": True,
         }
 
-    def test_gemini_pro_reasoning_clamps_to_supported_levels(self, transport):
+    def test_gemini_openai_compat_pro_reasoning_clamps_to_supported_levels(self, transport):
         msgs = [{"role": "user", "content": "Hi"}]
         kw = transport.build_kwargs(
             model="google/gemini-3.1-pro-preview",
             messages=msgs,
             provider_name="gemini",
+            base_url="https://generativelanguage.googleapis.com/v1beta/openai",
             reasoning_config={"enabled": True, "effort": "medium"},
         )
-        assert kw["extra_body"]["thinking_config"] == {
-            "includeThoughts": True,
-            "thinkingLevel": "low",
+        assert kw["extra_body"]["extra_body"]["google"]["thinking_config"] == {
+            "include_thoughts": True,
+            "thinking_level": "low",
         }
 
-    def test_gemini_disabled_reasoning_hides_thoughts(self, transport):
+    def test_gemini_native_disabled_reasoning_hides_thoughts(self, transport):
         msgs = [{"role": "user", "content": "Hi"}]
         kw = transport.build_kwargs(
             model="gemini-3-flash-preview",
             messages=msgs,
             provider_name="gemini",
+            base_url="https://generativelanguage.googleapis.com/v1beta",
             reasoning_config={"enabled": False},
         )
         assert kw["extra_body"]["thinking_config"] == {
             "includeThoughts": False,
         }
 
-    def test_gemini_xhigh_clamps_to_high(self, transport):
+    def test_gemini_openai_compat_xhigh_clamps_to_high(self, transport):
         msgs = [{"role": "user", "content": "Hi"}]
         kw = transport.build_kwargs(
             model="gemini-3-flash-preview",
             messages=msgs,
             provider_name="gemini",
+            base_url="https://generativelanguage.googleapis.com/v1beta/openai",
             reasoning_config={"enabled": True, "effort": "xhigh"},
         )
-        assert kw["extra_body"]["thinking_config"]["thinkingLevel"] == "high"
+        assert kw["extra_body"]["extra_body"]["google"]["thinking_config"]["thinking_level"] == "high"
+
+    def test_google_gemini_cli_keeps_top_level_thinking_config(self, transport):
+        msgs = [{"role": "user", "content": "Hi"}]
+        kw = transport.build_kwargs(
+            model="gemini-3-flash-preview",
+            messages=msgs,
+            provider_name="google-gemini-cli",
+            reasoning_config={"enabled": True, "effort": "high"},
+        )
+        assert kw["extra_body"]["thinking_config"] == {
+            "includeThoughts": True,
+            "thinkingLevel": "high",
+        }
+        assert "google" not in kw["extra_body"]
 
     def test_gemini_flash_minimal_clamps_to_low(self, transport):
         # Gemini 3 Flash documents low/medium/high; "minimal" isn't accepted,
@@ -199,11 +236,12 @@ def test_gemini_flash_minimal_clamps_to_low(self, transport):
             model="gemini-3-flash-preview",
             messages=msgs,
             provider_name="gemini",
+            base_url="https://generativelanguage.googleapis.com/v1beta/openai",
             reasoning_config={"enabled": True, "effort": "minimal"},
         )
-        assert kw["extra_body"]["thinking_config"] == {
-            "includeThoughts": True,
-            "thinkingLevel": "low",
+        assert kw["extra_body"]["extra_body"]["google"]["thinking_config"] == {
+            "include_thoughts": True,
+            "thinking_level": "low",
         }
 
     def test_max_tokens_with_fn(self, transport):

tests/run_agent/test_provider_parity.py

@@ -144,6 +144,36 @@ def test_strips_codex_only_tool_call_fields_from_chat_messages(self, monkeypatch
         assert messages[1]["tool_calls"][0]["response_item_id"] == "fc_123"
         assert "codex_reasoning_items" in messages[1]
 
+    def test_gemini_native_passes_base_url_for_top_level_thinking_config(self, monkeypatch):
+        agent = _make_agent(
+            monkeypatch,
+            "gemini",
+            base_url="https://generativelanguage.googleapis.com/v1beta",
+            model="gemini-3-flash-preview",
+        )
+        agent.reasoning_config = {"enabled": True, "effort": "high"}
+        kwargs = agent._build_api_kwargs([{"role": "user", "content": "hi"}])
+        assert kwargs["extra_body"]["thinking_config"] == {
+            "includeThoughts": True,
+            "thinkingLevel": "high",
+        }
+        assert "extra_body" not in kwargs["extra_body"]
+
+    def test_gemini_openai_compat_passes_base_url_for_nested_google_thinking_config(self, monkeypatch):
+        agent = _make_agent(
+            monkeypatch,
+            "gemini",
+            base_url="https://generativelanguage.googleapis.com/v1beta/openai",
+            model="gemini-3.1-pro-preview",
+        )
+        agent.reasoning_config = {"enabled": True, "effort": "high"}
+        kwargs = agent._build_api_kwargs([{"role": "user", "content": "hi"}])
+        assert "thinking_config" not in kwargs["extra_body"]
+        assert kwargs["extra_body"]["extra_body"]["google"]["thinking_config"] == {
+            "include_thoughts": True,
+            "thinking_level": "high",
+        }
+
     def test_should_sanitize_tool_calls_codex_vs_chat(self, monkeypatch):
         """Codex API should NOT sanitize, all other APIs should sanitize."""
         # Codex mode should NOT need sanitization

Closed issues · 1

Closed#16941[Bug]: Gemini chat-completions route does not bridge reasoning_config into thinking_config, so Discord show_reasoning stays empty

How this cluster was grouped

Link evidence (partly structural links): some members are connected by structural evidence (shared changed files, or a closing/superseded reference); the rest are joined by an automated relevance judgement over their text.
Members outside that structural core may belong to a different underlying problem.
Pairwise link density 0.29: the share of item pairs that carry a link, counting a structural link in full and a relevance-only link at 0.36. 1.00 means every pair is linked.

Pick up a Lab task

Check whether a fix covers the open issues here, with your own agent. Bring back evidence for maintainers.

Open the Lab

At a glance

Subsystem: agent
Merged PRs: #16999, #17605
Closed issues: 1
Attempts: 0 PRs closed without merging
Fix link checked: 29 Apr 2026

For your agent

The same facts, typed.

cluster/16941.json
cluster/16941.md