Merge pull request #947 from tnfssc/feat/kimi-k2

backend/agent/run.py

@@ -492,6 +492,9 @@ async def run_agent(
         elif "gemini-2.5-pro" in model_name.lower():
             # Gemini 2.5 Pro has 64k max output tokens
             max_tokens = 64000
+        elif "kimi-k2" in model_name.lower():
+            # Kimi-K2 has 120K context, set reasonable max output tokens
+            max_tokens = 8192
             
         generation = trace.generation(name="thread_manager.run_thread") if trace else None
         try:

backend/services/llm.py

@@ -232,6 +232,12 @@ def prepare_params(
     use_thinking = enable_thinking if enable_thinking is not None else False
     is_anthropic = "anthropic" in effective_model_name.lower() or "claude" in effective_model_name.lower()
     is_xai = "xai" in effective_model_name.lower() or model_name.startswith("xai/")
+    is_kimi_k2 = "kimi-k2" in effective_model_name.lower() or model_name.startswith("moonshotai/kimi-k2")
+
+    if is_kimi_k2:
+        params["provider"] = {
+            "order": ["groq", "together/fp8"]
+        }
 
     if is_anthropic and use_thinking:
         effort_level = reasoning_effort if reasoning_effort else 'low'

backend/utils/constants.py

@@ -53,6 +53,14 @@ MODELS = {
         },
         "tier_availability": ["paid"]
     },
+    "openrouter/moonshotai/kimi-k2": {
+        "aliases": ["moonshotai/kimi-k2", "kimi-k2"],
+        "pricing": {
+            "input_cost_per_million_tokens": 1.00,
+            "output_cost_per_million_tokens": 3.00
+        },
+        "tier_availability": ["paid"]
+    },
     "openai/gpt-4o": {
         "aliases": ["gpt-4o"],
         "pricing": {

frontend/src/components/thread/chat-input/_use-model-selection.ts

@@ -70,6 +70,12 @@ export const MODELS = {
     recommended: false,
     lowQuality: false
   },
+  'moonshotai/kimi-k2': { 
+    tier: 'premium', 
+    priority: 96,
+    recommended: false,
+    lowQuality: false
+  },
   'gpt-4.1': { 
     tier: 'premium', 
     priority: 96,