Crate menta 

Minimal Rust primitives for non-UI LLM and AI workflows.

menta focuses on a small core API for:

• text generation through provider-prefixed model ids such as openai/gpt-4.1-mini
• typed outputs with GenerateTextRequest<T>
• tool calling via #[derive(Tool)] and ToolExecute
• embeddings and similarity ranking

Quick Start

use menta::{GenerateTextRequest, generate_text};

#[tokio::main]
async fn main() -> Result<(), menta::Error> {
    let result = generate_text(
        GenerateTextRequest::new()
            .model("openai/gpt-4.1-mini")
            .prompt("Write a one-line summary of Rust."),
    )
    .await?;

    println!("{}", result.text);
    Ok(())
}

Typed Output

use menta::{GenerateTextRequest, generate_text};
use serde::Deserialize;

#[derive(Debug, Deserialize, schemars::JsonSchema)]
struct Summary {
    title: String,
    bullets: Vec<String>,
}

#[tokio::main]
async fn main() -> Result<(), menta::Error> {
    let result = generate_text(
        GenerateTextRequest::<Summary>::typed()
            .model("openai/gpt-4.1-mini")
            .prompt("Summarize Rust for backend engineers."),
    )
    .await?;

    println!("{}", result.output.title);
    Ok(())
}

Modules

providers

Macros

tools

Convenience macro for building a Vec<Tool> from one or more tool types.

Structs

BatchEmbeddingResult

Batched embedding result.

EmbeddingResult

Single embedding result.

GenerateTextRequest

Builder for text generation requests.

GenerateTextResult

Final result returned by generate_text.

ModelMessage

A provider-agnostic chat/message item used for prompts, history, and tool results.

ModelRequest

Low-level provider request produced from a GenerateTextRequest.

ModelResponse

Raw provider response before typed output parsing.

ModelSettings

Per-request model configuration.

ProviderRegistration

Provider registration entry collected through inventory.

Tool

Executable tool wrapper used in requests.

ToolCall

A single model-initiated tool call.

ToolDefinition

Public description of a tool exposed to a model.

ToolFieldSchema

A named field inside a ToolObjectSchema.

ToolObjectSchema

Object schema used by ToolSchema::Object.

ToolResult

Result produced after a tool call is executed.

Usage

Token usage summary returned by generation and embedding operations.

Enums

Error

Error type returned by the crate’s public APIs.

FinishReason

Reason a generation finished.

Output

Output mode used when parsing generation results.

Part

A content part inside a model message or model response.

Role

Model message role used in requests and responses.

StreamEvent

Event returned by stream_text.

ToolChoice

Controls whether the model may call tools.

ToolSchema

JSON-schema-like description used for tool inputs, tool outputs, and typed outputs.

Traits

EmbeddingModel

Trait implemented by embedding providers.

LanguageModel

Trait implemented by language model providers.

ToolExecute

Trait for executable tools that can be attached to a GenerateTextRequest.

ToolInput

Trait implemented by tool input types.

Functions

cosine_similarity

Computes cosine similarity for two embedding vectors.

embed

Embeds a single string with the given model id.

embed_many

Embeds multiple strings with the given model id.

generate_text

Generates text or structured output from a model.

language_model

Resolves a provider-prefixed model id to a LanguageModel.

provider_registry

Backwards-compatible alias for registered_providers.

rank_by_similarity

Scores and sorts named embeddings by similarity to the query embedding.

registered_providers

Returns the sorted set of registered provider ids.

stream_text

Runs a single provider stream pass and returns collected stream events.

tool

Creates an executable [Tool] from a type implementing ToolExecute.

Type Aliases

Result

Convenience result type used across the crate.

StreamTextStream

Derive Macros

Tool

Derive macro for declaring tool input schemas from a struct with named fields.