Luke Harwood

Posted on Feb 24 • Updated on Mar 2

openai-proxz - An intuitive OpenAI library for zig!

#showcase #library

https://github.com/lukeharwood11/openai-proxz

I wanted a simple interface for interacting with OpenAI & compatible APIs and couldn't find one that was MIT licensed and carried the features I needed, so I built one!

As someone who is coming from python, I loved how simple the openai-python package was, so this was modeled after that interface.

📙 ProxZ Docs: https://proxz.mle.academy

Features

Built-in retry logic
Environment variable config support for API keys, org. IDs, project IDs, and base urls
Integration with the most popular OpenAI endpoints with a generic request method for missing endpoints
Streamed responses
Logging configurability

Installation

To install proxz, run

 zig fetch --save "git+https://github.com/lukeharwood11/openai-proxz"

And add the following to your build.zig

const proxz = b.dependency("proxz", .{
    .target = target,
    .optimize = optimize,
});

exe.root_module.addImport("proxz", proxz.module("proxz"));

Usage

Client Configuration

const proxz = @import("proxz");
const OpenAI = proxz.OpenAI;

// make sure you have an OPENAI_API_KEY environment variable set,
// or pass in a .api_key field to explicitly set!
var openai = try OpenAI.init(allocator, .{});
defer openai.deinit();

Chat Completions

Regular

const ChatMessage = proxz.ChatMessage;

var response = try openai.chat.completions.create(.{
    .model = "gpt-4o",
    .messages = &[_]ChatMessage{
        .{
            .role = "user",
            .content = "Hello, world!",
        },
    },
});
// This will free all the memory allocated for the response
defer response.deinit();
std.log.debug("{s}", .{response.choices[0].message.content});

Streamed Response

var stream = try openai.chat.completions.createStream(.{
    .model = "gpt-4o-mini",
    .messages = &[_]ChatMessage{
        .{
            .role = "user",
            .content = "Write me a poem about lizards. Make it a paragraph or two.",
        },
    },
});
defer stream.deinit();

std.debug.print("\n", .{});
while (try stream.next()) |val| {
    std.debug.print("{s}", .{val.choices[0].delta.content});
}
std.debug.print("\n", .{});

Embeddings

const inputs = [_][]const u8{ "Hello", "Foo", "Bar" };
const response = try openai.embeddings.create(.{
    .model = "text-embedding-3-small",
    .input = &inputs,
});
// Don't forget to free resources!
defer response.deinit();
std.log.debug("Model: {s}\nNumber of Embeddings: {d}\nDimensions of Embeddings: {d}", .{
    response.model,
    response.data.len,
    response.data[0].embedding.len,
});

Models

Get model details

var response = try openai.models.retrieve("gpt-4o");
defer response.deinit();
std.log.debug("Model is owned by '{s}'", .{response.owned_by});

List all models

var response = try openai.models.list();
defer response.deinit();
std.log.debug("The first model you have available is '{s}'", .{response.data[0].id})

Configuring Logging

By default all logs are enabled for your entire application.
To configure your application, and set the log level for proxz, include the following in your main.zig.

pub const std_options = std.Options{
    .log_level = .debug, // this sets your app level log config
    .log_scope_levels = &[_]std.log.ScopeLevel{
        .{
            .scope = .proxz,
            .level = .info, // set to .debug, .warn, .info, or .err
        },
    },
};

All logs in proxz use the scope .proxz, so if you don't want to see debug/info logs of the requests being sent, set .level = .err. This will only display when an error occurs that proxz can't recover from.

Contributions

Contributions are welcome and encouraged! Submit an issue for any bugs/feature requests and open a PR if you tackled one of them!

Zig NEWS