Add function calling to ai-proxy (#3727)

Author: tomek-labuk

app/_data/plugins/ai-proxy.yaml

@@ -23,6 +23,13 @@ providers:
       route_type: 'llm/v1/embeddings'
       model_example: '<a href="https://docs.aws.amazon.com/bedrock/latest/userguide/model-ids.html">Use the model name for the specific LLM provider</a>'
       min_version: '3.11'
+    function_calling:
+      supported: true
+      streaming: false
+      upstream_path: 'Uses the <code>Converse</code> API with tool configuration'
+      route_type: 'llm/v1/chat'
+      model_example: 'Model-dependent. Supported for Claude, Command, and select models'
+      min_version: '3.8'
     batches: # Native format from SDK only
       supported: 'n/a'
       streaming: false
@@ -82,6 +89,13 @@ providers:
       route_type: 'llm/v1/completions'
       model_example: 'claude-2.1'
       min_version: '3.6'
+    function_calling:
+      supported: true
+      streaming: false
+      upstream_path: '/v1/messages'
+      route_type: 'llm/v1/chat'
+      model_example: 'claude-3-opus-20240229'
+      min_version: '3.6'
     batches: # Native format from SDK only
       supported: 'n/a'
       streaming: true
@@ -116,6 +130,13 @@ providers:
       route_type: 'llm/v1/embeddings'
       model_example: 'text-embedding-ada-002<sup>1</sup>'
       min_version: '3.11'
+    function_calling:
+      supported: true
+      streaming: false
+      upstream_path: '/openai/deployments/{deployment_name}/chat/completions'
+      route_type: 'llm/v1/chat'
+      model_example: 'gpt-4'
+      min_version: '3.6'
     files:
       supported: true
       streaming: false
@@ -232,6 +253,13 @@ providers:
       route_type: 'llm/v1/embeddings'
       model_example: 'embed-english-v3.0'
       min_version: '3.11'
+    function_calling:
+      supported: true
+      streaming: false
+      upstream_path: '/v1/chat'
+      route_type: 'llm/v1/chat'
+      model_example: 'command-r-plus'
+      min_version: '3.6'
 
   - name: 'Dashscope'
     url_pattern: 'https://dashscope.aliyuncs.com or https://dashscope-intl.aliyuncs.com'
@@ -283,6 +311,13 @@ providers:
       route_type: 'llm/v1/embeddings'
       model_example: 'text-embedding-004'
       min_version: '3.11'
+    function_calling:
+      supported: true
+      streaming: false
+      upstream_path: 'Uses <code>generateContent</code> API with function declarations'
+      route_type: 'llm/v1/chat'
+      model_example: 'gemini-2.0-flash'
+      min_version: '3.8'
     files: # Native format from SDK only
       supported: 'n/a'
       streaming: false
@@ -358,6 +393,13 @@ providers:
       route_type: 'llm/v1/embeddings'
       model_example: 'text-embedding-004'
       min_version: '3.11'
+    function_calling:
+      supported: true
+      streaming: false
+      upstream_path: 'Uses <code>generateContent</code> API with function declarations'
+      route_type: 'llm/v1/chat'
+      model_example: 'gemini-2.0-flash'
+      min_version: '3.8'
     files:
       supported: 'n/a'
       streaming: false
@@ -468,6 +510,13 @@ providers:
       route_type: 'llm/v1/embeddings'
       model_example: 'mistral-embed'
       min_version: '3.11'
+    function_calling:
+      supported: true
+      streaming: false
+      upstream_path: '/v1/chat/completions or user-defined'
+      route_type: 'llm/v1/chat'
+      model_example: 'mistral-large-latest'
+      min_version: '3.6'
 
   - name: 'OpenAI'
     formats: 'GPT-3.5, GPT-4, GPT-4o, and Multi-Modal'
@@ -494,6 +543,13 @@ providers:
       route_type: 'llm/v1/embeddings'
       model_example: 'text-embedding-ada-002<sup>1</sup>'
       min_version: '3.11'
+    function_calling:
+      supported: true
+      streaming: false
+      upstream_path: '/v1/chat/completions'
+      route_type: 'llm/v1/chat'
+      model_example: 'gpt-4'
+      min_version: '3.6'
     files:
       supported: true
       streaming: false
@@ -591,6 +647,13 @@ providers:
     embeddings:
       supported: false
       streaming: false
+    function_calling:
+      supported: true
+      streaming: false
+      upstream_path: '/v1/chat/completions'
+      route_type: 'llm/v1/chat'
+      model_example: 'grok-2-latest'
+      min_version: '3.13'
     files:
       supported: false
       streaming: false
@@ -637,4 +700,4 @@ parameters:
   route_type: 'config.targets.route_type'
   options: 'config.targets[].model.options'
   upstream_url: 'config.targets[].model.options.upstream_url'
-  model_name: 'config.targets[].model.name'
+  model_name: 'config.targets[].model.name'

app/_includes/plugins/ai-proxy/overview.md

@@ -205,10 +205,13 @@ Support for chat, completions, and embeddings:
 
 ### Advanced text generation {% new_in 3.11 %}
 
-Support for function calling, tool use, and batch processing:
+Support for files and batch processing and function calling (tool use):
 
 {% include plugins/ai-proxy/tables/supported-providers-processing.html providers=providers %}
 
+{:.info}
+> Function calling uses the llm/v1/chat route type.
+
 ### Audio features {% new_in 3.11 %}
 
 Support for text-to-speech, transcription, and translation:

app/_includes/plugins/ai-proxy/tables/supported-providers-processing.html

@@ -1,6 +1,6 @@
 {% assign note_counter = 0 %}
 {% assign notes = "" | split: "" %}
-{% assign features = "files,batches,assistants,responses" | split: "," %}
+{% assign features = "files,batches,assistants,responses,function_calling" | split: "," %}
 
 {% comment %}First pass: collect all notes{% endcomment %}
 {% for provider in include.providers %}
@@ -24,6 +24,7 @@
       <th>Batches</th>
       <th>Assistants</th>
       <th>Responses</th>
+      <th>Function Calling</th>
     </tr>
   </thead>
   <tbody>