fix(kimi): resolve Kimi Code API 403 errors and User-Agent restrictions (#3549)

* fix(kimi): resolve Kimi Code API 403 errors and User-Agent restrictions

Kimi Code subscription keys require a whitelisted coding-agent User-Agent to avoid access_terminated_error 403s. This adds User-Agent probing and caching for Kimi Code endpoints.

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(kimi): omit temperature for kimi-for-coding API calls

Kimi Code rejects any non-default temperature with HTTP 400, which broke deep research probes and low-temp LLM rounds.

Co-authored-by: Cursor <cursoragent@cursor.com>

---------

Co-authored-by: Cursor <cursoragent@cursor.com>

tests/test_kimi_code_hosts.py

@@ -0,0 +1,32 @@
+"""Kimi Code host-allowlist behavior (follow-up to provider support).
+
+Kimi Code (https://api.kimi.com/coding/v1) is a subscription, OpenAI-compatible
+cloud API with native tool-calling. These tests pin the three host-list integrations:
+  - agent loop sends native tool schemas to Kimi Code (not fenced-block parsing),
+  - teacher escalation treats Kimi Code as SOTA (loop OFF, no added latency).
+"""
+from src import agent_loop, teacher_escalation
+
+
+class TestAgentToolHosts:
+    def test_kimi_code_in_api_hosts(self):
+        assert "api.kimi.com" in agent_loop._API_HOSTS
+
+    def test_kimi_code_url_matches_api_host(self):
+        url = "https://api.kimi.com/coding/v1/chat/completions"
+        assert any(h in url for h in agent_loop._API_HOSTS)
+
+    def test_unknown_host_not_matched(self):
+        url = "https://example.invalid/v1/chat/completions"
+        assert not any(h in url for h in agent_loop._API_HOSTS)
+
+
+class TestTeacherEscalationSota:
+    def test_kimi_code_is_sota_not_self_hosted(self):
+        assert teacher_escalation.is_self_hosted("https://api.kimi.com/coding/v1/chat/completions") is False
+
+    def test_known_cloud_still_sota(self):
+        assert teacher_escalation.is_self_hosted("https://api.openai.com/v1") is False
+
+    def test_local_endpoint_still_self_hosted(self):
+        assert teacher_escalation.is_self_hosted("http://localhost:8000/v1") is True

tests/test_kimi_code_user_agent.py

@@ -0,0 +1,69 @@
+"""Kimi Code User-Agent fallback list and 403 detection."""
+from src.llm_core import (
+    KIMI_CODE_USER_AGENTS,
+    KIMI_CODE_USER_AGENT,
+    _is_kimi_code_access_denied,
+    _is_kimi_code_url,
+    _kimi_code_base_key,
+    _kimi_code_ua_cache,
+    _kimi_code_ua_candidates,
+    _remember_kimi_code_user_agent,
+    httpx_post_kimi_aware,
+)
+
+
+class TestKimiCodeUserAgents:
+    def test_default_is_first_fallback(self):
+        assert KIMI_CODE_USER_AGENT == KIMI_CODE_USER_AGENTS[0]
+
+    def test_multiple_fallbacks_configured(self):
+        assert len(KIMI_CODE_USER_AGENTS) >= 3
+        assert "KimiCLI/1.0" in KIMI_CODE_USER_AGENTS
+
+    def test_detects_coding_agent_403(self):
+        body = '{"error":{"message":"only available for Coding Agents","type":"access_terminated_error"}}'
+        assert _is_kimi_code_access_denied(403, body) is True
+
+    def test_non_403_not_access_denied(self):
+        assert _is_kimi_code_access_denied(401, "unauthorized") is False
+
+    def test_ua_candidates_prefers_cache(self):
+        _kimi_code_ua_cache.clear()
+        url = "https://api.kimi.com/coding/v1/chat/completions"
+        _remember_kimi_code_user_agent(url, "Kilo-Code/1.0")
+        candidates = _kimi_code_ua_candidates(url)
+        assert candidates[0] == "Kilo-Code/1.0"
+        assert len(candidates) == len(KIMI_CODE_USER_AGENTS)
+        _kimi_code_ua_cache.clear()
+
+    def test_non_kimi_url_has_no_candidates(self):
+        assert _kimi_code_ua_candidates("https://api.openai.com/v1") == []
+
+    def test_base_key_normalizes_chat_url(self):
+        assert _kimi_code_base_key("https://api.kimi.com/coding/v1/chat/completions") == (
+            "https://api.kimi.com/coding/v1"
+        )
+
+    def test_post_retries_next_user_agent_on_403(self, monkeypatch):
+        _kimi_code_ua_cache.clear()
+        calls = []
+
+        class _Resp:
+            def __init__(self, status, text=""):
+                self.status_code = status
+                self.content = text.encode()
+                self.text = text
+
+        def fake_post(url, headers=None, **kwargs):
+            calls.append(headers.get("User-Agent"))
+            if headers.get("User-Agent") == KIMI_CODE_USER_AGENTS[0]:
+                return _Resp(403, '{"error":{"type":"access_terminated_error"}}')
+            return _Resp(200, "{}")
+
+        monkeypatch.setattr("src.llm_core.httpx.post", fake_post)
+        url = "https://api.kimi.com/coding/v1/chat/completions"
+        r = httpx_post_kimi_aware(url, {"Authorization": "Bearer x"}, json={})
+        assert r.status_code == 200
+        assert calls[0] == KIMI_CODE_USER_AGENTS[0]
+        assert calls[1] == KIMI_CODE_USER_AGENTS[1]
+        _kimi_code_ua_cache.clear()

tests/test_llm_core_temperature.py

@@ -14,7 +14,7 @@ from src import llm_core
 @pytest.mark.parametrize(
     "model",
     ["o1", "o1-mini", "o3", "o3-mini", "o4-mini", "gpt-5", "gpt-5-mini",
-     "openrouter/openai/o3-mini", "OpenAI/GPT-5"],
+     "openrouter/openai/o3-mini", "OpenAI/GPT-5", "kimi-for-coding"],
 )
 def test_reasoning_models_restrict_temperature(model):
     assert llm_core._restricts_temperature(model) is True
@@ -62,6 +62,12 @@ def test_reasoning_model_payload_omits_temperature(monkeypatch):
     assert payload["max_completion_tokens"] == 5
 
 
+def test_kimi_for_coding_payload_omits_temperature(monkeypatch):
+    payload = _capture_openai_payload(monkeypatch, "kimi-for-coding", 0.1)
+    assert "temperature" not in payload
+    assert payload["max_tokens"] == 5
+
+
 def test_normal_model_payload_keeps_temperature(monkeypatch):
     payload = _capture_openai_payload(monkeypatch, "gpt-4o", 0.2)
     assert payload["temperature"] == 0.2

tests/test_model_routes.py

@@ -205,6 +205,9 @@ class TestMatchProviderCurated:
     def test_ollama_url(self):
         assert _match_provider_curated("https://ollama.com/api", "openai") == "ollama"
 
+    def test_kimi_code_url(self):
+        assert _match_provider_curated("https://api.kimi.com/coding/v1", "openai") == "kimi-code"
+
     def test_no_url_match_returns_provider(self):
         assert _match_provider_curated("https://localhost:1234", "openai") == "openai"
 
@@ -312,6 +315,12 @@ class TestCurateModels:
         assert curated == models
         assert extra == []
 
+    def test_kimi_code_partitions(self):
+        models = ["kimi-for-coding", "other-model"]
+        curated, extra = _curate_models(models, "kimi-code")
+        assert "kimi-for-coding" in curated
+        assert "other-model" in extra
+
     def test_curated_sorted_by_priority(self):
         models = ["gpt-4o-mini", "gpt-4o", "o3"]
         curated, _ = _curate_models(models, "openai")