Gemini changes for core client, fiddle, and docs

docs/docs/syntax/client/client.mdx

@@ -34,8 +34,8 @@ BAML ships with the following providers (you can can also write your own!):
   - `openai`
   - `azure-openai`
   - `anthropic`
-  - `ollama`
   - `google-ai`
+  - `ollama`
 - Composite client providers
   - `fallback`
   - `round-robin`
@@ -111,6 +111,21 @@ client<llm> MyClient {
   }
 }
 ```
+### Google 
+
+Provider names:
+- `google-ai`
+
+Accepts any options as defined by the [Gemini SDK](https://ai.google.dev/gemini-api/docs/get-started/tutorial?lang=rest#configuration).
+
+```rust
+client<llm> MyGoogleClient {
+  provider google-ai
+  options{
+    model "gemini-1.5-pro-001"
+  }
+}
+```
 
 ### Ollama
 
@@ -135,6 +150,7 @@ client<llm> MyOllamaClient {
 
 1. For Ollama, in your terminal run `ollama serve`
 2. In another window, run `ollama run llama2` (or your model), and you should be good to go.
+3. If your Ollama port is not 11434, you can specify the endpoint manually.
 
 ```rust
 client<llm> MyClient {
@@ -143,28 +159,14 @@ client<llm> MyClient {
     model llama2
     options {
       temperature 0
+      base_url "http://localhost:<ollama_port>" // Default is 11434
     }
   }
 }
 ```
 
-### Google 
-
-Provider names:
-- `google-ai`
-
-Accepts any options as defined by the [Gemini SDK](https://ai.google.dev/gemini-api/docs/api-overview).
 
-
-```rust
-client<llm> MyGoogleClient {
-  provider google-ai
-  options{
-    model "gemini-1.5-pro-001"
-    api_key env.GOOGLE_API_KEY
-  }
-}
-```
+This is not the Vertex AI Gemini API, but the Google Generative AI Gemini API, which supports the same models but at a different endpoint.
 
 
 ### Fallback

engine/baml-runtime/src/internal/llm_client/primitive/google/google_client.rs

@@ -267,19 +267,15 @@ impl RequestBuilder for GoogleClient {
         prompt: either::Either<&String, &Vec<RenderedChatMessage>>,
         stream: bool,
     ) -> reqwest::RequestBuilder {
-        let mut should_stream = "generateContent?";
+        let mut should_stream = "generateContent";
         if stream {
-            should_stream = "streamGenerateContent?alt=sse&";
+            should_stream = "streamGenerateContent?alt=sse";
         }
 
         let baml_original_url = format!(
-            "https://generativelanguage.googleapis.com/v1/models/{}:{}key={}",
+            "https://generativelanguage.googleapis.com/v1/models/{}:{}",
             self.properties.model_id.as_ref().unwrap_or(&"".to_string()),
-            should_stream,
-            self.properties
-                .api_key
-                .clone()
-                .unwrap_or_else(|| "".to_string())
+            should_stream
         );
 
         let mut req = self.client.post(
@@ -295,6 +291,13 @@ impl RequestBuilder for GoogleClient {
         }
 
         req = req.header("baml-original-url", baml_original_url);
+        req = req.header(
+            "x-goog-api-key",
+            self.properties
+                .api_key
+                .clone()
+                .unwrap_or_else(|| "".to_string()),
+        );
 
         let mut body = json!(self.properties.properties);
         let body_obj = body.as_object_mut().unwrap();

typescript/fiddle-frontend/package.json

@@ -3,7 +3,7 @@
   "version": "0.1.0",
   "private": true,
   "scripts": {
-    "dev": "infisical run --env=test -- next dev",
+    "dev": "next dev",
     "build": "next build",
     "start": "next start",
     "lint": "next lint"

typescript/fiddle-frontend/public/_examples/clients.baml

@@ -1,4 +1,6 @@
-// These are LLM clients you can use in your functions. We currently support Anthropic, OpenAI / Azure, and Ollama as providers but are expanding to many more.
+// These are LLM clients you can use in your functions. We currently support Anthropic, OpenAI / Azure, Gemini, and Ollama as providers but are expanding to many more.
+
+// We also support any other provider that follows the OpenAI API specification, such as HuggingFace.
 
 // For this playground, we have setup a few clients for you to use already with some free credits.
 
@@ -36,4 +38,12 @@ client<llm> Claude {
     max_tokens 1000
 
   }
+}
+
+client<llm> Gemini {
+  provider google-ai
+  options{
+    model "gemini-1.5-pro-001"
+    api_key env.GOOGLE_API_KEY
+  }
 }

typescript/fiddle-frontend/public/_examples/intro/chat-roles/baml_src/clients.baml

@@ -1,4 +1,6 @@
-// These are LLM clients you can use in your functions. We currently support Anthropic, OpenAI / Azure, and Ollama as providers but are expanding to many more.
+// These are LLM clients you can use in your functions. We currently support Anthropic, OpenAI / Azure, Gemini, and Ollama as providers but are expanding to many more.
+
+// We also support any other provider that follows the OpenAI API specification, such as HuggingFace.
 
 // For this playground, we have setup a few clients for you to use already with some free credits.
 
@@ -36,4 +38,13 @@ client<llm> Claude {
     max_tokens 1000
 
   }
-}
+}
+
+client<llm> Gemini {
+  provider google-ai
+  options{
+    model "gemini-1.5-pro-001"
+    api_key env.GOOGLE_API_KEY
+  }
+}
+

typescript/fiddle-frontend/public/_examples/intro/classify-message/baml_src/clients.baml

@@ -1,4 +1,6 @@
-// These are LLM clients you can use in your functions. We currently support Anthropic, OpenAI / Azure, and Ollama as providers but are expanding to many more.
+// These are LLM clients you can use in your functions. We currently support Anthropic, OpenAI / Azure, Gemini, and Ollama as providers but are expanding to many more.
+
+// We also support any other provider that follows the OpenAI API specification, such as HuggingFace.
 
 // For this playground, we have setup a few clients for you to use already with some free credits.
 
@@ -36,4 +38,13 @@ client<llm> Claude {
     max_tokens 1000
 
   }
-}
+}
+
+client<llm> Gemini {
+  provider google-ai
+  options{
+    model "gemini-1.5-pro-001"
+    api_key env.GOOGLE_API_KEY
+  }
+}
+

typescript/fiddle-frontend/public/_examples/intro/extract-resume/baml_src/clients.baml

@@ -1,4 +1,6 @@
-// These are LLM clients you can use in your functions. We currently support Anthropic, OpenAI / Azure, and Ollama as providers but are expanding to many more.
+// These are LLM clients you can use in your functions. We currently support Anthropic, OpenAI / Azure, Gemini, and Ollama as providers but are expanding to many more.
+
+// We also support any other provider that follows the OpenAI API specification, such as HuggingFace.
 
 // For this playground, we have setup a few clients for you to use already with some free credits.
 
@@ -36,4 +38,13 @@ client<llm> Claude {
     max_tokens 1000
 
   }
-}
+}
+
+client<llm> Gemini {
+  provider google-ai
+  options{
+    model "gemini-1.5-pro-001"
+    api_key env.GOOGLE_API_KEY
+  }
+}
+

typescript/fiddle-frontend/public/_examples/intro/images/baml_src/clients.baml

@@ -1,4 +1,6 @@
-// These are LLM clients you can use in your functions. We currently support Anthropic, OpenAI / Azure, and Ollama as providers but are expanding to many more.
+// These are LLM clients you can use in your functions. We currently support Anthropic, OpenAI / Azure, Gemini, and Ollama as providers but are expanding to many more.
+
+// We also support any other provider that follows the OpenAI API specification, such as HuggingFace.
 
 // For this playground, we have setup a few clients for you to use already with some free credits.
 
@@ -36,4 +38,12 @@ client<llm> Claude {
     max_tokens 1000
 
   }
-}
+}
+
+client<llm> Gemini {
+  provider google-ai
+  options{
+    model "gemini-1.5-pro-001"
+    api_key env.GOOGLE_API_KEY
+  }
+}

typescript/fiddle-proxy/.gitignore

@@ -1 +1,2 @@
-node_modules
+node_modules
+.env

typescript/fiddle-proxy/package.json

@@ -11,11 +11,14 @@
   "author": "",
   "license": "ISC",
   "devDependencies": {
-    "@flydotio/node-demo": "^0.2.1"
+    "@flydotio/node-demo": "^0.2.1",
+    "cors": "^2.8.5"
+
   },
   "dependencies": {
     "cors": "^2.8.5",
     "express": "^4.19.2",
+    "dotenv": "16.4.5",
     "http-proxy-middleware": "^3.0.0"
   }
 }

typescript/fiddle-proxy/server.js

@@ -1,12 +1,20 @@
 const cors = require('cors')
 const { createProxyMiddleware } = require('http-proxy-middleware')
 const app = require('express')()
+require('dotenv').config()
 
 app.use(cors())
 
 app.use(
   createProxyMiddleware({
     changeOrigin: true,
+    pathRewrite: (path, req) => {
+      // Ensure the URL does not end with a slash
+      if (path.endsWith('/')) {
+        return path.slice(0, -1)
+      }
+      return path
+    },
     router: (req) => {
       // Extract the original target URL from the custom header
       const originalUrl = req.headers['baml-original-url']
@@ -32,6 +40,12 @@ app.use(
           }
           proxyReq.setHeader('x-api-key', process.env.ANTHROPIC_API_KEY)
         }
+        if (req.headers['baml-original-url'].includes('gemini')) {
+          if (process.env.GOOGLE_API_KEY === undefined) {
+            throw new Error('GOOGLE_API_KEY is missing')
+          }
+          proxyReq.setHeader('x-goog-api-key', process.env.GOOGLE_API_KEY)
+        }
       },
       proxyRes: (proxyRes, req, res) => {
         proxyRes.headers['Access-Control-Allow-Origin'] = '*'