This OCaml project is a AI enhanced developer toolkit that integrates with OpenAI's API. Key features include code indexing and natural language search, prompt generation, and a markdown style ai chat interface that includes enhanced functionality using openai's function calling. This library contains alot of experimentation, and is intented to be more of a research project than a production ready app. Uses Effect based library Eio for concurrency and Owl for computing cosign similarity of openai embeddings. I recommmend installing in a local opam switch.
To build this project, you can use OPAM:
opam install . --deps-only
# If owl has issues installing on new apple chips https://github.com/owlbarn/owl/issues/597#issuecomment-1119470934
# try this
opam pin -n git+https://github.com/mseri/owl.git#arm64 --with-version=1.1.0
PKG_CONFIG_PATH="/opt/homebrew/opt/openblas/lib/pkgconfig" opam install owl.1.1.0This command-line application is designed for indexing OCaml code, serving queries to a code vector search database using OpenAI embeddings, and interacting with gpt models through chat interface using OpenAI chat completion api. The application provides four main commands: index, query, chat-completion, and tokenize.
The index command indexes OCaml code in a specified folder for a code vector search database using OpenAI embeddings.
index -folder-to-index <folder_path> -vector-db-folder <db_folder_path>-folder-to-index: Path to the folder containing OCaml code to index. Default is./lib.-vector-db-folder: Path to the folder to store vector database data. Default is./vector.
The query command queries the indexed OCaml code using natural language.
query -vector-db-folder <db_folder_path> -query-text <query_text> -num-results <num_results>-vector-db-folder: Path to the folder containing vector database data. Default is./vector.-query-text: Natural language query text to search the indexed OCaml code. This is a required parameter.-num-results: Number of top results to return. Default is 5.
The chat-completion command calls the OpenAI API to provide chat completion based on the content of a prompt file.
chat-completion -prompt-file <prompt_file_path> -output-file <output_file_path> -max-tokens <max_tokens>-prompt-file: Path to the file containing the initial prompt. This is optional. If provided, its content is loaded and appended to the output file.-output-file: Path to the file to save the chat completion output. Default is./prompts/default.md. Only include this parameter and not -prompt-file if you wan to continue with a previous conversation-max-tokens: Maximum number of tokens to generate in the chat completion. Default is 600.
The markup syntax used is designed to represent a conversation as a series of messages. Each message is represented as a msg element with attributes for the role of the message (e.g., "system", "user", "assistant", "function"), an optional name to represent messages from the user/assistant or name of the function called for results returned in a msg with role function, and an optional function_call when the assistant invokes a function_call. The content of the message is represented as the text content of the msg element. If a function call is present, it is represented as a function_call attribute with a function_name attribute and arguments set as the elements contents. Only msg elements with assistant role can have a function_call. The function name and arguments are used to call a function from the list of functions made availble to the prompt. You can view availible functions in lib/chat_completion.ml or just ask the model in the chat for a list of availble function. The results of a function call are put in a msg element with role function, name is the function used, and the contents of the element are the results.
here's an example of a conversation using the specialized markup:
<msg role="system">You are a helpful assistant.</msg>
<msg role="user" name="John">What's the weather like?</msg>
<msg role="assistant">I'm not sure, would you like me to look it up for you?</msg>
<msg role="user" name="John">Yes, please.</msg>
<msg role="assistant" function_call function_name="get_url_content">{"url": "http://api.weatherapi.com/v1/current.json?key=YOUR_API_KEY&q=London"}</msg>
<msg role="function" name="get_url_content">{"location":{"name":"London","region":"City of London, Greater London","country":"UK"},"current":{"temp_c":14.0,"condition":{"text":"Partly cloudy"}}}</msg>
<msg role="assistant">The current weather in London is partly cloudy with a temperature of 14 degrees Celsius.</msg>
<msg role="user" name="John">Thank you!</msg>
<msg role="assistant">You're welcome!</msg>The tokenize command tokenizes the provided file using the OpenAI Tikitoken spec.
tokenize -file <file_path>-file: Path to the file to tokenize. Default isbin/main.ml.
Indexing a folder:
index -folder-to-index ./my_ocaml_code -vector-db-folder ./my_vector_dbQuerying the indexed code:
query -vector-db-folder ./my_vector_db -query-text "How to write a for loop in OCaml?" -num-results 10Running a chat completion:
chat-completion -prompt-file ./my_prompt.md -output-file ./my_chat.md -max-tokens 1000Tokenizing a file:
tokenize -file ./my_ocaml_code/my_file.mlnote: If testing using dune prefix commands with
dune exe ./bin/main.exe --Any contributions you make are greatly appreciated.
If you have a suggestion that would make this better, please fork the repo and create a pull request. You can also simply open an issue with the tag "enhancement".
Distributed under the MIT License. See LICENSE.txt for more information.
This project is Highly Experimental