feat!: Use Mercury's OpenAI client (#7)

* WIP: added OpenAI.V1 in OpenAICompatible

* WIP: removed OpenAI internal module

* WIP: fixed examples as per new OpenAI client

* WIP: added messageConvertible instance for OpenAI.Message

* chore!: fixed examples with new openai client

* chore: doc fixs

Author: tusharad

docs/docs/concepts/DocumentLoader.md

@@ -41,7 +41,7 @@ Right now, langchain-hs provides below integrations, with more integrations plan
 - `PdfLoader`: Loads documents from a PDF file path.
 - `DirectoryLoader`: Loads documents from a directory. It can recursively load files from subdirectories and filter files based on their extensions.
 
-**DiretoryLaoderOptions**
+**DirectoryLoaderOptions**
 
 ```haskell
 data DirectoryLoaderOptions = DirectoryLoaderOptions

docs/docs/concepts/Embeddings.md

@@ -50,11 +50,11 @@ data DeepseekEmbedding = DeepseekEmbedding {
   model :: Text
 }
 
-instance Embeddings Deepseek where
-  embedDocuments OpenAIEmbedding docs = do
+instance Embeddings DeepseekEmbedding where
+  embedDocuments DeepseekEmbedding{..} docs = do
     -- Your implementation here
     return $ Right []
-  embedQuery OpenAIEmbedding{..} query = do
+  embedQuery DeepseekEmbedding{..} query = do
     -- Your implementation here
     return $ Right []
 ```

docs/docs/concepts/LLM.md

@@ -68,7 +68,7 @@ data Message = Message
 The `Message` type represents a single message in the chat conversation. It has three fields:
   - `role`: The role of the message sender (Possible values are User, Assistant, System, Tool, Developer, Function).
   - `content`: The content of the message (Text).
-  - `messageData`: Additional data associated with the message, such as name or toolcalls (soon to be depreceted). You can use `defaultMessageData` to create a default instance of `MessageData` with no additional data.
+  - `messageData`: Additional data associated with the message, such as name or tool calls. You can use `defaultMessageData` to create a default instance of `MessageData` with no additional data.
 
 ## Writing your own LLM
 
@@ -158,7 +158,7 @@ For full control, construct the type directly:
 let customClient = OpenAICompatible
   { apiKey = Just "your-api-key"
   , modelName = "custom-mistral"
-  , callbacks = [stdOutCallbacl]  -- Example with logging callback
+  , callbacks = [stdOutCallback]  -- Example with logging callback
   , baseUrl = Just "https://my-ai-server.com/openai/v1"
   , defaultBaseUrl = "http://localhost:8080/v1"
   , providerName = "MyCustomProvider"

docs/docs/concepts/Tool.md

@@ -172,7 +172,7 @@ Calculator tool can perform Add, Sub, Mul, Div, Pow operation and return type wo
 
 :::Warning
 
-Write now, duckduckgo tool only returns result if the search term has an abstract card. 
+Right now, duckduckgo tool only returns result if the search term has an abstract card. 
 It does not return links, only abstract summary
 
 :::

docs/docs/concepts/VectorStore.md

@@ -29,7 +29,7 @@ The `InMemory` vector store is a simple implementation that stores documents and
 Below is an example that:
  - Reads a 20 pages long pdf file and generates a list of documents.
  - Sets up an OllamaEmbeddings model.
- - Creates an InMemoryVectoStore with model and documents.
+ - Creates an InMemoryVectorStore with model and documents.
  - Fetches top 2 documents that are closest to the given query.
 
  ```haskell

docs/docs/getting-started/quick_start.md

@@ -79,16 +79,16 @@ main :: IO ()
 main = do
   let openAI = OpenAI
         { apiKey = "your-api-key"
-        , openAIModelName = "gpt-4.1-nano"
         , callbacks = []
+        , baseUrl = Nothing
         }
   result <- LLM.generate openAI "Tell me a joke" Nothing
   case result of
     Left err -> putStrLn $ "Error: " ++ err
     Right response -> putStrLn response
     ```
 In above code:
-1. Setup the `OpenAI` LLM with the api key, model name and optional list of callback functions.
+1. Setup the `OpenAI` LLM with the api key and optional list of callback functions. You can specify a custom base URL if needed.
 2. Call the `generate` function with the prompt and optional parameters.
 3. Handle the result, which can be either an error or the generated text.
 4. The `generate` function returns a `Text` response, which you can print or use as needed.
@@ -154,7 +154,7 @@ runApp :: IO ()
 runApp = do
   let ollamaLLM = Ollama "llama3.2" []  
   let chatHistory = fromList
-        [ Message System "Explain everthing with a texas accent." defaultMessageData
+        [ Message System "Explain everything with a texas accent." defaultMessageData
         , Message User "What is functional programming?" defaultMessageData
         ]
   chatResult <- chat ollamaLLM chatHistory Nothing
@@ -177,7 +177,7 @@ For Ollama, Make sure the model that you want to use is installed on your local
 `Message` constructor takes 3 parameters:
 1. `role`: The role of the message sender (System, User, Assistant).
 2. `content`: The content of the message (Text).
-3. `metadata`: Optional metadata for the message (A type containing a optinal name and optional list of toolnames) (Currently unstable).
+3. `messageData`: Optional metadata for the message (A type containing an optional name and optional list of tool calls).
 4. `defaultMessageData`: A default value for the metadata, which can be used if no specific metadata is provided.
 
 `chat` takes a `NonEmpty` list of `Message` as input. The `NonEmpty` type ensures that the list is not empty, which is important for chat history. 
@@ -200,7 +200,6 @@ main :: IO ()
 main = do
   let openAI = OpenAI
         { apiKey = "your-api-key"
-        , openAIModelName = "gpt-4.1-nano"
         , callbacks = []
         , baseUrl = Nothing
         }
@@ -214,7 +213,7 @@ main = do
     Right response -> putStrLn $ "Chat Response:\n" ++ T.unpack response
     ```
 In above code:
-1. Setup the `OpenAI` LLM with the api key, model name and optional list of callback functions.
+1. Setup the `OpenAI` LLM with the api key and optional list of callback functions. You can specify a custom base URL if needed.
 2. Create a `chatHistory` using `fromList` with `Message` constructor.
 3. Call the `chat` function with the `chatHistory` and optional parameters.
 4. Handle the result, which can be either an error or the generated text.
@@ -224,7 +223,7 @@ In above code:
 `Message` constructor takes 3 parameters:
 1. `role`: The role of the message sender (System, User, Assistant).
 2. `content`: The content of the message (Text).
-3. `metadata`: Optional metadata for the message (A type containing a optinal name and optional list of toolnames) (Currently unstable).
+3. `metadata`: Optional metadata for the message (A type containing a optional name and optional list of toolnames) (Currently unstable).
 4. `defaultMessageData`: A default value for the metadata, which can be used if no specific metadata is provided.
 `chat` takes a `NonEmpty` list of `Message` as input. The `NonEmpty` type ensures that the list is not empty, which is important for chat history.
 :::
@@ -270,7 +269,7 @@ In above code:
 `Message` constructor takes 3 parameters:
 1. `role`: The role of the message sender (System, User, Assistant).
 2. `content`: The content of the message (Text).
-3. `metadata`: Optional metadata for the message (A type containing a optinal name and optional list of toolnames) (Currently unstable).
+3. `metadata`: Optional metadata for the message (A type containing a optional name and optional list of toolnames) (Currently unstable).
 4. `defaultMessageData`: A default value for the metadata, which can be used if no specific metadata is provided.
 `chat` takes a `NonEmpty` list of `Message` as input. The `NonEmpty` type ensures that the list is not empty, which is important for chat history.
 :::
@@ -332,7 +331,7 @@ The StreamHandler takes two functions:
 `Message` constructor takes 3 parameters:
 1. `role`: The role of the message sender (System, User, Assistant).
 2. `content`: The content of the message (Text).
-3. `metadata`: Optional metadata for the message (A type containing a optinal name and optional list of toolnames) (Currently unstable).
+3. `metadata`: Optional metadata for the message (A type containing a optional name and optional list of toolnames) (Currently unstable).
 4. `defaultMessageData`: A default value for the metadata, which can be used if no specific metadata is provided.
 `stream` takes a `NonEmpty` list of `Message` as input. The `NonEmpty` type ensures that the list is not empty, which is important for chat history.
 :::
@@ -355,7 +354,6 @@ runApp = do
     let openAI =
             OpenAI
                 { apiKey = "your-api-key"
-                , openAIModelName = "gpt-4.1-nano"
                 , callbacks = []
                 , baseUrl = Nothing
                 }
@@ -391,7 +389,7 @@ The StreamHandler takes two functions:
 `Message` constructor takes 3 parameters:
 1. `role`: The role of the message sender (System, User, Assistant).
 2. `content`: The content of the message (Text).
-3. `metadata`: Optional metadata for the message (A type containing a optinal name and optional list of toolnames) (Currently unstable).
+3. `metadata`: Optional metadata for the message (A type containing a optional name and optional list of toolnames) (Currently unstable).
 4. `defaultMessageData`: A default value for the metadata, which can be used if no specific metadata is provided.
 `stream` takes a `NonEmpty` list of `Message` as input. The `NonEmpty` type ensures that the list is not empty, which is important for chat history.
 :::
@@ -449,7 +447,7 @@ The StreamHandler takes two functions:
 `Message` constructor takes 3 parameters:
 1. `role`: The role of the message sender (System, User, Assistant).
 2. `content`: The content of the message (Text).
-3. `metadata`: Optional metadata for the message (A type containing a optinal name and optional list of toolnames) (Currently unstable).
+3. `metadata`: Optional metadata for the message (A type containing a optional name and optional list of toolnames) (Currently unstable).
 4. `defaultMessageData`: A default value for the metadata, which can be used if no specific metadata is provided.
 `stream` takes a `NonEmpty` list of `Message` as input. The `NonEmpty` type ensures that the list is not empty, which is important for chat history.
 :::

docs/docs/index.md

@@ -4,14 +4,14 @@ sidebar_position: 1
 
 # langchain-hs Documentation
 
-Lanchain-hs is the Haskell implementation of [LangChain](https://www.langchain.com/).
+Langchain-hs is the Haskell implementation of [LangChain](https://www.langchain.com/).
 
 ## Welcome to langchain-hs
 
 Langchain-hs brings the power of LangChain to Haskell, enabling type-safe LLM application development with:
 
 - Composable pipelines using Haskell's type system
-- Closely following with Offical LangChain concepts
+- Closely following with Official LangChain concepts
 - Performance through functional programming patterns
 - Support for Ollama, OpenAI and custom LLM integrations
 
@@ -21,7 +21,7 @@ Check out the [Quickstart](getting-started/quick_start) guide to get started wit
 
 ## Components 
 
-This documenation provides an overview of the core components of langchain-hs. Each component is designed to be modular, composable and interchangeble, allowing you to build complex applications with ease. 
+This documentation provides an overview of the core components of langchain-hs. Each component is designed to be modular, composable and interchangeable, allowing you to build complex applications with ease. 
 
 Langchain-hs is built around the core concepts of LangChain; You will find documentation for each components including:
 

docs/docusaurus.config.ts

@@ -91,7 +91,7 @@ const config: Config = {
           items: [
             {
               label: 'Github',
-              to: 'https://github.com/tusharad/lanchain-hs',
+              to: 'https://github.com/tusharad/langchain-hs',
             },
           ],
         },

examples/langchain-examples/app/Main.hs

@@ -3,16 +3,16 @@ module Main (main) where
 -- import App.OpenRouter.PdfQA
 -- import App.OpenRouter.ToolCall
 -- import App.Ollama.StructuredOutput
--- import App.OpenRouter.StructuredOutput
+import App.OpenRouter.StructuredOutput
+
 -- import App.Ollama.MultiAgent
 -- import App.Ollama.Agent
 -- import App.Gemini.Embeddings
 -- import App.Gemini.Runnable
-import App.Gemini.ImageInput
-
+-- import App.Gemini.ImageInput
 -- import App.Gemini.Simple
-
 -- import App.Gemini.Streaming
+-- import App.OpenAI.ImageInput
 
 main :: IO ()
 main = runApp

examples/langchain-examples/langchain-examples.cabal

@@ -53,9 +53,11 @@ library
     , filepath
     , langchain-hs
     , ollama-haskell
+    , openai
     , scientific
     , text
     , transformers
+    , vector
   default-language: Haskell2010
 
 executable langchain-examples-exe
@@ -76,9 +78,11 @@ executable langchain-examples-exe
     , langchain-examples
     , langchain-hs
     , ollama-haskell
+    , openai
     , scientific
     , text
     , transformers
+    , vector
   default-language: Haskell2010
 
 test-suite langchain-examples-test
@@ -100,7 +104,9 @@ test-suite langchain-examples-test
     , langchain-examples
     , langchain-hs
     , ollama-haskell
+    , openai
     , scientific
     , text
     , transformers
+    , vector
   default-language: Haskell2010