Merge d2e2b9b3d5 into 3caab85931

aider/models.py

@@ -1,3 +1,4 @@
+import asyncio
 import difflib
 import hashlib
 import importlib.resources
@@ -15,6 +16,7 @@ import json5
 import yaml
 from PIL import Image
 
+from aider import ollama
 from aider.dump import dump  # noqa: F401
 from aider.llm import litellm
 from aider.openrouter import OpenRouterModelManager
@@ -994,6 +996,16 @@ def register_models(model_settings_fnames):
 
 def register_litellm_models(model_fnames):
     files_loaded = []
+
+    # Add available ollama models
+    if os.getenv("OLLAMA_API_BASE"):
+        try:
+            model_def = asyncio.run(ollama.query_available_models())
+            model_info_manager.local_model_metadata.update(model_def)
+        except Exception as e:
+            raise Exception(f"Error querying ollama models: {e}")
+
+    # Load from static model database
     for model_fname in model_fnames:
         if not os.path.exists(model_fname):
             continue

aider/ollama.py

@@ -0,0 +1,69 @@
+import asyncio
+import os
+
+import aiohttp
+
+
+async def query_available_models():
+    api_base = os.getenv("OLLAMA_API_BASE")
+    if not api_base:
+        return {}
+
+    async with aiohttp.ClientSession() as session:
+        # Ping the tags endpoint to get model names
+        async with session.get(f"{api_base}/api/tags") as response:
+            if response.status != 200:
+                return {}
+
+            tags = await response.json()
+            model_names = [tag["name"] for tag in tags["models"]]
+
+            # Wait for all model descriptions to complete
+            model_descriptions = await asyncio.gather(
+                *[describe_ollama_model(model_name) for model_name in model_names]
+            )
+
+            # Merge the results into a single dictionary
+            result = {}
+            for model_desc in model_descriptions:
+                result.update(model_desc)
+
+            return result
+
+
+async def describe_ollama_model(model_name):
+    api_base = os.getenv("OLLAMA_API_BASE")
+    context_length = None
+    async with aiohttp.ClientSession() as session:
+        # Ping the /show endpoint to get context length
+        async with session.post(f"{api_base}/api/show", json={"model": model_name}) as response:
+            if response.status != 200:
+                return {}
+            json = await response.json()
+            model_info = json.get("model_info")
+            for key in model_info:
+                # Model native context length is usually stored in a key like
+                # "llama.context_length" or "qwen3.context_length"
+                if "context_length" in key:
+                    context_length = model_info[key]
+                    break
+
+    return {
+        "ollama/"
+        + model_name: {
+            "max_tokens": context_length,
+            "max_input_tokens": context_length,
+            "max_output_tokens": context_length,
+            "input_cost_per_token": 0,
+            "input_cost_per_token_cache_hit": 0,
+            "cache_read_input_token_cost": 0,
+            "cache_creation_input_token_cost": 0,
+            "output_cost_per_token": 0,
+            "litellm_provider": "ollama",
+            "mode": "chat",
+            "supports_function_calling": False,
+            "supports_assistant_prefill": False,
+            "supports_tool_choice": False,
+            "supports_prompt_caching": False,
+        }
+    }