Crate neuron_provider_ollama 

neuron-provider-ollama

Ollama provider for the neuron agent blocks ecosystem. Implements the Provider trait from neuron-types against the Ollama Chat API, supporting both synchronous completions and newline-delimited JSON (NDJSON) streaming.

Ollama runs locally and requires no API key or authentication. The default model is llama3.2. The default base URL is http://localhost:11434.

Installation

cargo add neuron-provider-ollama

Key Types

• Ollama – client struct with builder methods (new, from_env, model, base_url, keep_alive). Implements Provider from neuron-types and Default.
• ProviderError – re-exported error type for all provider failures.
• StreamHandle – returned by complete_stream, yields StreamEvent items as the model generates tokens.

Features

• No authentication required – designed for local Ollama instances.
• keep_alive control for model memory residency ("5m", "0" to unload immediately).
• Tool call support using the OpenAI-compatible format that Ollama adopted.
• NDJSON streaming parsed line-by-line from the response byte stream.
• Tool call IDs synthesized via uuid::Uuid::new_v4() since Ollama does not provide them natively.
• Tool support varies by model — not all Ollama models support function calling. Check the Ollama model library for which models support tools.

Usage

use neuron_provider_ollama::Ollama;
use neuron_types::{CompletionRequest, Message, Provider};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Default constructor (no auth needed for local Ollama)
    let provider = Ollama::new()
        .model("llama3.2")
        .keep_alive("5m");

    // Or from environment (reads OLLAMA_HOST if set, defaults to localhost:11434)
    let provider = Ollama::from_env()?;

    let request = CompletionRequest {
        messages: vec![Message::user("What is the capital of France?")],
        max_tokens: Some(256),
        ..Default::default()
    };

    let response = provider.complete(request).await?;

    for block in &response.message.content {
        println!("{block:?}");
    }
    Ok(())
}

Error handling

Each provider defines its own ProviderError type. If you’re using multiple providers, pattern-match on the specific provider’s error rather than expecting a shared error type across providers.

Part of neuron

This crate is one block in the neuron composable agent toolkit. It depends only on neuron-types.

License

Licensed under either of Apache License, Version 2.0 or MIT License at your option.

Re-exports

pub use client::Ollama;

Modules

client

Ollama API client struct and builder.

mapping

Request/response mapping between neuron-types and the Ollama Chat API format.

Structs

StreamHandle

Handle to a streaming completion response.

Enums

ProviderError

Errors from LLM provider operations.

StreamEvent

An event emitted during streaming completion.