Skip to content

The fastest, cheapest, most efficient open-source agent harness. Run more than 500 agents on a $10 VPS.

merge: integrate inference epic with latest canonical base

Steven Enamakeled9a0530eeef
diff --cc crates/openhuman-core/src/inference/readiness_tests.rs
index 03fa87dbcb,3372f5ce95..08b36ab193
--- a/crates/openhuman-core/src/inference/readiness_tests.rs
+++ b/crates/openhuman-core/src/inference/readiness_tests.rs
@@@ -4,58 -3,104 +3,103 @@@ use tempfile::TempDir
#[tokio::test]
async fn cached_probe_serves_definitive_negative_results() {
+ use crate::core::runtime::{CoreContext, DispatchCtx, DomainSet};
+ use crate::core::slots::Extensions;
+ use crate::inference::registry::{register_chat_model_in, ChatModelEntry, ProviderModelHooks};
use std::sync::atomic::{AtomicUsize, Ordering};
+ use std::sync::Arc;
+
+ let _model_override_guard =
+ crate::inference::provider::factory::test_provider_override::exclusive();
let tmp = TempDir::new().unwrap();
- let mut config = Config {
+ let config = Config {
workspace_dir: tmp.path().join("workspace"),
action_dir: tmp.path().join("workspace"),
config_path: tmp.path().join("config.toml"),
..Config::default()
};
- AuthService::new(tmp.path(), false)
- .store_provider_token(
- APP_SESSION_PROVIDER,
- DEFAULT_AUTH_PROFILE_NAME,
- "test.session.jwt",
- std::collections::HashMap::new(),
- true,
- )
- .expect("seed app-session token");
let hit_count = std::sync::Arc::new(AtomicUsize::new(0));
+ // The core owns no hosted provider; the managed backend's readiness probe
+ // is whatever the host registered. Register one that rejects the way the
+ // backend does, and count how often the core actually asks it.
+ let _guard = crate::inference::inference_test_guard_async().await;
+ let _exclusive = crate::inference::provider::factory::test_provider_override::exclusive();
let counter = hit_count.clone();
- let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap();
- let addr = listener.local_addr().unwrap();
- let app = axum::Router::new().route(
- "/openai/v1/chat/completions",
- axum::routing::post(move || {
- let counter = counter.clone();
- async move {
+ let managed = crate::inference::registry::chat_model(
+ crate::inference::provider::factory::PROVIDER_OPENHUMAN,
+ )
+ .expect("the unit-test managed provider is available");
+ crate::inference::registry::register_chat_model(
+ crate::inference::provider::factory::PROVIDER_OPENHUMAN,
+ crate::inference::registry::ChatModelEntry {
+ build: managed.build,
+ probe: Some(std::sync::Arc::new(move |_role, _model, _config| {
counter.fetch_add(1, Ordering::SeqCst);
- use axum::response::IntoResponse;
- (
- axum::http::StatusCode::BAD_REQUEST,
- axum::Json(serde_json::json!({
- "success": false,
- "error": "API key not configured for provider",
- "errorCode": "BAD_REQUEST"
- })),
- )
- .into_response()
- }
- }),
+ Box::pin(async { Err("API key not configured for provider".to_string()) })
+ })),
+ },
);
- tokio::spawn(async move { axum::serve(listener, app).await.unwrap() });
- config.api_url = Some(format!("http://{addr}"));
- let first = cached_probe(&DispatchCtx::ambient(), "summarization", &config).await;
+ let extensions = Extensions::new();
+ register_chat_model_in(
+ &extensions,
+ crate::inference::provider::factory::PROVIDER_OPENHUMAN,
+ ChatModelEntry {
+ build: Arc::new(|_| {
+ Ok((
+ Arc::new(tinyagents_harness::testkit::ScriptedModel::replies(
+ ["ready"].to_vec(),
+ )),
+ "fixture-model".to_string(),
+ ))
+ }),
+ probe: Some(Arc::new(|_, _, config| {
+ Box::pin(async move {
+ let base = config.api_url.as_deref().unwrap().trim_end_matches('/');
+ let response = reqwest::Client::new()
+ .post(format!("{base}/openai/v1/chat/completions"))
+ .json(&serde_json::json!({"model":"fixture-model","messages":[]}))
+ .send()
+ .await
+ .map_err(|error| error.to_string())?;
+ let status = response.status();
+ let body = response.text().await.map_err(|error| error.to_string())?;
+ if status.is_success() {
+ Ok(())
+ } else {
+ Err(body)
+ }
+ })
+ })),
+ models: Some(ProviderModelHooks {
+ default_model: Arc::new(|| Some("fixture-model".to_string())),
+ role_for_model: Arc::new(|_| None),
+ is_alias: Arc::new(|_| false),
+ is_raw_model: Arc::new(|model| !model.trim().is_empty()),
+ supports_vision: Arc::new(|_| false),
+ resolve_model: Arc::new(|_, _, _| Some("fixture-model".to_string())),
+ context_window: Arc::new(|_| None),
+ default_pricing: Arc::new(|| None),
+ without_reasoning: Arc::new(tinyinference_llm::model::without_reasoning),
+ }),
+ },
+ );
+ let core = CoreContext::for_test_with_extensions(
+ DomainSet::full(),
+ Some(config.workspace_dir.clone()),
+ extensions,
+ );
+ let ctx = DispatchCtx::new(core);
+
+ let first = cached_probe(&ctx, "summarization", &config).await;
assert!(first
.unwrap_err()
.to_ascii_lowercase()
.contains("api key not configured for provider"));
assert_eq!(hit_count.load(Ordering::SeqCst), 1);
- assert!(
- cached_probe(&DispatchCtx::ambient(), "summarization", &config)
- .await
- .is_err()
- );
+ assert!(cached_probe(&ctx, "summarization", &config).await.is_err());
assert_eq!(hit_count.load(Ordering::SeqCst), 1);
+ crate::inference::registry::unregister_chat_model(
+ crate::inference::provider::factory::PROVIDER_OPENHUMAN,
+ );
}