feat(llm): send an explicit reasoning_effort for Kimi models that support it (GRAPHIFY_KIMI_EFFORT)

CHANGELOG.md

@@ -2,6 +2,10 @@
 
 Full release notes with details on each version: [GitHub Releases](https://github.com/safishamsi/graphify/releases)
 
+## Unreleased
+
+- Feature: the `kimi` backend now sends an explicit `reasoning_effort` (default `max`, overridable via `GRAPHIFY_KIMI_EFFORT`) for Kimi models that support it — K3 advertises `valid_efforts ["low","high","max"]` on `/models`, and sending nothing let the server default (`"high"` for K3) apply silently while the gemini backend already carried an effort setting. Forwarded by the existing request plumbing; no behaviour change for models that ignore the field.
+
 ## 0.9.49 (2026-08-24)
 
 - Feature: `graphify merge-graphs` now links a type declaration that two repos share — same fully-qualified namespace and name, from different repos — with a `same_type_as` edge, so a shared contract type is navigable across the repo boundary; two unrelated types that merely share a short name are not linked (#3007, thanks @durmazoguzhan).

graphify/llm.py

@@ -123,6 +123,12 @@ def _resolve_ollama_base_url(default: str) -> str:
         "pricing": {"input": 0.74, "output": 4.66},  # USD per 1M tokens
         "temperature": None,  # kimi-k2.6 enforces its own fixed temperature; sending any value raises 400
         "max_tokens": 16384,
+        # Reasoning effort for Kimi models that support it (K3 advertises
+        # valid_efforts ["low","high","max"] on /models). Sending nothing lets the
+        # server default ("high" for K3) apply silently, while the gemini block
+        # above carries an effort setting. Forwarded as `reasoning_effort` by the
+        # existing request plumbing; models that ignore the field are unaffected.
+        "reasoning_effort": os.environ.get("GRAPHIFY_KIMI_EFFORT", "max"),
     },
     "ollama": {
         "base_url": _resolve_ollama_base_url("http://localhost:11434/v1"),

tests/test_kimi_reasoning_effort.py

@@ -0,0 +1,31 @@
+"""Tests for the kimi backend's reasoning_effort config (GRAPHIFY_KIMI_EFFORT).
+
+Kimi K3 advertises valid_efforts ["low","high","max"] on /models; sending
+nothing let the server default ("high") apply silently. The backend config now
+carries an explicit effort, defaulting to "max", overridable via env — the
+same import-time pattern as ANTHROPIC_BASE_URL on the claude backend.
+"""
+
+import importlib
+
+from graphify import llm
+
+
+def test_kimi_reasoning_effort_defaults_to_max(monkeypatch):
+    monkeypatch.delenv("GRAPHIFY_KIMI_EFFORT", raising=False)
+    reloaded = importlib.reload(llm)
+    try:
+        assert reloaded.BACKENDS["kimi"]["reasoning_effort"] == "max"
+    finally:
+        monkeypatch.undo()
+        importlib.reload(llm)
+
+
+def test_kimi_reasoning_effort_env_override(monkeypatch):
+    monkeypatch.setenv("GRAPHIFY_KIMI_EFFORT", "low")
+    reloaded = importlib.reload(llm)
+    try:
+        assert reloaded.BACKENDS["kimi"]["reasoning_effort"] == "low"
+    finally:
+        monkeypatch.undo()
+        importlib.reload(llm)