Files
owlen/docs/provider-implementation.md

99 lines
3.5 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Provider Implementation Guide
This guide explains how to implement a new provider for Owlen. Providers are the components that connect to different LLM APIs.
## The `ModelProvider` Trait
The core of the provider system is the `ModelProvider` trait, located in `owlen-core::provider`. Any new provider must implement this async trait so it can be managed by `ProviderManager`.
Here is a simplified version of the trait:
```rust
use async_trait::async_trait;
use owlen_core::provider::{GenerateChunk, GenerateRequest, GenerateStream, ModelInfo, ProviderMetadata, ProviderStatus};
#[async_trait]
pub trait ModelProvider: Send + Sync {
fn metadata(&self) -> &ProviderMetadata;
async fn health_check(&self) -> owlen_core::Result<ProviderStatus>;
async fn list_models(&self) -> owlen_core::Result<Vec<ModelInfo>>;
async fn generate_stream(&self, request: GenerateRequest) -> owlen_core::Result<GenerateStream>;
}
```
## Creating a New Crate
1. **Create a new crate** in the `crates/` directory. For example, `owlen-myprovider`.
2. **Add dependencies** to your new crate's `Cargo.toml`. You will need `owlen-core`, `async-trait`, `tokio`, and any crates required for interacting with the new API (e.g., `reqwest`).
3. **Add the new crate to the workspace** in the root `Cargo.toml`.
## Implementing the Trait
In your new crate's `lib.rs`, you will define a struct for your provider and implement the `Provider` trait for it.
```rust
use async_trait::async_trait;
use owlen_core::provider::{
GenerateRequest, GenerateStream, ModelInfo, ModelProvider, ProviderMetadata,
ProviderStatus, ProviderType,
};
pub struct MyProvider {
metadata: ProviderMetadata,
client: MyHttpClient,
}
impl MyProvider {
pub fn new(config: &MyConfig) -> owlen_core::Result<Self> {
let metadata = ProviderMetadata::new(
"my_provider",
"My Provider",
ProviderType::Cloud,
true,
);
Ok(Self {
metadata,
client: MyHttpClient::new(config)?,
})
}
}
#[async_trait]
impl ModelProvider for MyProvider {
fn metadata(&self) -> &ProviderMetadata {
&self.metadata
}
async fn health_check(&self) -> owlen_core::Result<ProviderStatus> {
self.client.ping().await.map(|_| ProviderStatus::Available)
}
async fn list_models(&self) -> owlen_core::Result<Vec<ModelInfo>> {
self.client.list_models().await
}
async fn generate_stream(&self, request: GenerateRequest) -> owlen_core::Result<GenerateStream> {
self.client.generate(request).await
}
}
```
## Integrating with Owlen
Once your provider is implemented, you will need to register it with the `ProviderManager` and surface it to users.
1. **Add your provider crate** as a dependency to the component that will host it (an MCP server or `owlen-cli`).
2. **Register the provider** with `ProviderManager` during startup:
```rust
let manager = ProviderManager::new(config);
manager.register_provider(Arc::new(MyProvider::new(config)?)).await;
```
3. **Update configuration docs/examples** so the provider has a `[providers.my_provider]` entry.
4. **Expose via MCP (optional)** if the provider should run out-of-process. Owlens TUI talks to providers exclusively via MCP after Phase 10.
5. **Add tests** similar to `crates/owlen-providers/tests/integration_test.rs` that exercise registration, model aggregation, generation routing, and health transitions.
For concrete examples, see the Ollama providers in `crates/owlen-providers/` and the integration tests added in commit 13.