7 releases
0.0.7 | Aug 28, 2024 |
---|---|
0.0.6 | Jul 28, 2024 |
0.0.4 | Jun 25, 2024 |
#123 in Machine learning
76 downloads per month
Used in misan
48KB
804 lines
Misanthropy
Misanthropy is set of Rust bindings for Anthropic API, providing easy access to Claude and other Anthropic models. It consists of two main components:
misanthropy
: A Rust client library for the Anthropic APImisan
: A command-line interface (CLI) tool for quick interactions with the API
Features
- Simple, idiomatic Rust interface for the Anthropic API
- Support for text and image content in messages
- Support for streaming real-time responses
- Configurable client with defaults for model and token limits
- CLI tool for quick interactions with the API from the command line
Usage
Library
Here's a basic example of using the misanthropy
library:
use misanthropy::{Anthropic, MessagesRequest, Content};
#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
let client = Anthropic::from_env()?;
let mut request = MessagesRequest::default();
request.add_user(Content::text("Hello, Claude!"));
let response = client.messages(request).await?;
println!("{}", response.format_nicely());
Ok(())
}
For more examples, please check the examples
directory in the misanthropy
crate. These examples demonstrate various
features and use cases of the library.
CLI
The misan
CLI tool provides a command-line interface for interacting with the
Anthropic API. For usage instructions, run:
misan --help
Configuration
ANTHROPIC_API_KEY
: Set this environment variable with your Anthropic API key.- Default model and max tokens can be set when creating the
Anthropic
client or overridden per request.
Advanced Features
Streaming Responses
The library supports streaming responses for real-time interactions:
let mut stream = client.messages_stream(request)?;
while let Some(event) = stream.next().await {
match event {
Ok(event) => {
// Handle the streaming event
}
Err(e) => eprintln!("Error: {}", e),
}
}
Advanced Features
Using Tools
The library supports defining and using tools in conversations. Tools are
defined using the schemars
crate to generate JSON schemas for the tool
inputs.
- First, add
schemars
to your dependencies:
[dependencies]
schemars = "0.8"
- Define your tool input structure and derive
JsonSchema
:
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
/// Get the current weather for a location.
#[derive(JsonSchema, Serialize, Deserialize)]
struct GetWeather {
/// The city and country, e.g., "London, UK"
location: String,
/// Temperature unit: "celsius" or "fahrenheit"
unit: Option<String>,
}
- Create a
Tool
from your input structure:
use misanthropy::{Anthropic, MessagesRequest, Tool};
let weather_tool = Tool::new::<GetWeather>();
- Add the tool to your request:
let request = MessagesRequest::default()
.with_tool(weather_tool)
.with_system(vec![Content::text("You can use the GetWeather tool to check the weather.")]);
- When the AI uses the tool, you can deserialize the input:
if let Some(tool_use) = response.content.iter().find_map(|content| {
if let Content::ToolUse(tool_use) = content {
Some(tool_use)
} else {
None
}
}) {
if tool_use.name == "GetWeather" {
let weather_input: GetWeather = serde_json::from_value(tool_use.input.clone())?;
println!("Weather requested for: {}", weather_input.location);
// Here you would typically call an actual weather API
}
}
This approach allows you to define strongly-typed tool inputs that the AI can use, while also providing a way to handle the tool usage in your code.
License
This project is licensed under the MIT License - see the LICENSE file for details.
Dependencies
~7–18MB
~250K SLoC