feat(harness): add host-composable turn and progress APIs

README.md

@@ -177,7 +177,8 @@ changing the outer transcript or treating a preview as a write.
 ## Subagent orchestration
 
 `tinyagents-orchestration` owns child-agent composition and the shared
-`SubAgentJobRegistry`. The `subagents` feature in `tinyagents-harness-tools`
+`tinyagents_orchestration::subagent::SubAgentJobRegistry`. The `subagents`
+feature in `tinyagents-harness-tools`
 exposes `tinyagents_harness_tools::subagents`, which provides
 `DriverSubAgentTool`, the host-planned typed dispatcher: its `background` mode
 starts a child and returns a stable job id, while `inline` mode waits for the

crates/tinyagents-graph/Cargo.toml

@@ -17,7 +17,7 @@ tinyagents-tasks = { path = "../tinyagents-tasks", version = "2.2.0", default-fe
 tinyagents-harness = { path = "../tinyagents-harness", version = "2.2.0", default-features = false, features = [
     "langfuse",
 ] }
-tinyinference-llm = { path = "../../vendor/tinyinference/crates/tinyinference-llm", version = "0.4.1" }
+tinyinference-llm = { path = "../../vendor/tinyinference/crates/tinyinference-llm", version = "0.5.0" }
 tinytools = { path = "../../vendor/tinytools/crates/tinytools", version = "0.6.0" }
 tokio = { workspace = true, features = ["sync", "time", "macros", "rt", "rt-multi-thread", "fs"] }
 tracing = { workspace = true }

crates/tinyagents-harness-tools/Cargo.toml

@@ -16,7 +16,7 @@ serde = { workspace = true }
 serde_json = { workspace = true }
 tinyagents-harness = { path = "../tinyagents-harness", version = "2.2.0", default-features = false }
 tinyagents-tasks = { path = "../tinyagents-tasks", version = "2.2.0", default-features = false, optional = true }
-tinyinference-llm = { path = "../../vendor/tinyinference/crates/tinyinference-llm", version = "0.4.1", optional = true }
+tinyinference-llm = { path = "../../vendor/tinyinference/crates/tinyinference-llm", version = "0.5.0", optional = true }
 tinytools = { path = "../../vendor/tinytools/crates/tinytools", version = "0.6.0" }
 tinyagents-orchestration = { path = "../tinyagents-orchestration", version = "2.2.0", default-features = false, optional = true }
 tokio = { workspace = true, features = ["sync", "time", "macros", "rt", "rt-multi-thread"] }

crates/tinyagents-harness-tools/README.md

@@ -25,8 +25,8 @@ run-state middleware and its required `goals` and `todos` modules.
 - `skills` provides `UseSkillTool`, `PackCatalog`, `PackRegistryHandle`, and
   `UseSkillDispatch` for binding a selected tool from a named pack.
 - `subagents` provides `DriverSubAgentTool`, `SubAgentJobsTool`, and
-  `SubAgentMessageTool` over the orchestration crate's shared
-  `SubAgentJobRegistry`.
+  `SubAgentMessageTool` over the registry owned by
+  `tinyagents_orchestration::subagent::SubAgentJobRegistry`.
 - `thread_locks` provides the public `ThreadLockMap` concurrency primitive.
 - The crate root exports `CurrentTimeTool`, `ResolveTimeTool`,
   `AskClarificationTool`, `PlanExitTool`, and the `per_call` binding adapter.

crates/tinyagents-harness-tools/src/subagents/mod.rs

@@ -4,7 +4,6 @@ mod driver_tool;
 mod jobs;
 
 pub use driver_tool::{DriverSubAgentTool, SubagentDriverToolHost};
-pub use tinyagents_orchestration::subagent::SubAgentJobRegistry;
 #[cfg(test)]
 #[path = "driver_tool_tests.rs"]
 mod driver_tool_tests;

crates/tinyagents-harness-tools/src/subagents/sessions_tests.rs

@@ -1,9 +1,9 @@
-use crate::subagents::{SubAgentJobAction, SubAgentJobRegistry, SubAgentJobsTool};
+use crate::subagents::{SubAgentJobAction, SubAgentJobsTool};
 use std::sync::Arc;
 use tinyagents_harness::context::{RunConfig, RunContext};
 use tinyagents_harness::ids::new_call_id;
 use tinyagents_harness::tool::ToolDispatch;
-use tinyagents_orchestration::subagent::*;
+use tinyagents_orchestration::subagent::{SubAgentJobRegistry, *};
 use tinyagents_tasks::TaskStore;
 use tinystoragedrivers_core::{DocumentStore, MemoryStorage, Scope, StorageBackend};
 
@@ -213,7 +213,7 @@ async fn jobs_tool_overlays_scoped_sessions_and_awaits_user_envelopes() {
             .with_lineage(parent_run.clone(), parent_run.clone()),
         )
         .unwrap();
-    let jobs = SubAgentJobsTool::new(crate::subagents::SubAgentJobRegistry::default())
+    let jobs = SubAgentJobsTool::new(SubAgentJobRegistry::default())
         .with_sessions(Arc::new(store.clone()), |owner| owner.to_string())
         .with_task_store(task_store, move |_| parent_run.clone());
 
@@ -356,7 +356,7 @@ async fn jobs_wait_recovers_terminal_results_from_the_durable_task_store() {
         )
         .unwrap();
 
-    let jobs = SubAgentJobsTool::new(crate::subagents::SubAgentJobRegistry::default())
+    let jobs = SubAgentJobsTool::new(SubAgentJobRegistry::default())
         .with_fixed_action(crate::subagents::SubAgentJobAction::Await)
         .with_identifier_fields("task_id", "subagent_session_id")
         .with_task_store(task_store.clone(), move |_| owner.clone())
@@ -410,7 +410,7 @@ async fn jobs_wait_recovers_terminal_results_from_the_durable_task_store() {
     assert_eq!(failed["error"], "durable failure");
     assert_eq!(failed["agent"], "reviewer");
 
-    let foreign_jobs = SubAgentJobsTool::new(crate::subagents::SubAgentJobRegistry::default())
+    let foreign_jobs = SubAgentJobsTool::new(SubAgentJobRegistry::default())
         .with_fixed_action(crate::subagents::SubAgentJobAction::Await)
         .with_identifier_field("task_id")
         .with_task_store(task_store.clone(), |_| {

crates/tinyagents-harness/Cargo.toml

@@ -31,11 +31,12 @@ thiserror = "2"
 tracing = { workspace = true }
 tinyagents-definition = { path = "../tinyagents-definition", version = "2.2.0" }
 tinytools-agent = { path = "../../vendor/tinytools/crates/tinytools-agent", version = "0.6.0", default-features = false }
-tinyinference-llm = { path = "../../vendor/tinyinference/crates/tinyinference-llm", version = "0.4.1" }
-tinyinference-core = { path = "../../vendor/tinyinference/crates/tinyinference-core", version = "0.4.1" }
-tinyinference-embeddings = { path = "../../vendor/tinyinference/crates/tinyinference-embeddings", version = "0.4.1" }
-tinyinference-image = { path = "../../vendor/tinyinference/crates/tinyinference-image", version = "0.4.1", optional = true }
-tinyinference-video = { path = "../../vendor/tinyinference/crates/tinyinference-video", version = "0.4.1", optional = true }
+tinyinference-llm = { path = "../../vendor/tinyinference/crates/tinyinference-llm", version = "0.5.0" }
+tinyinference-providers = { path = "../../vendor/tinyinference/crates/tinyinference-providers", version = "0.5.0" }
+tinyinference-core = { path = "../../vendor/tinyinference/crates/tinyinference-core", version = "0.5.0" }
+tinyinference-embeddings = { path = "../../vendor/tinyinference/crates/tinyinference-embeddings", version = "0.5.0" }
+tinyinference-image = { path = "../../vendor/tinyinference/crates/tinyinference-image", version = "0.5.0", optional = true }
+tinyinference-video = { path = "../../vendor/tinyinference/crates/tinyinference-video", version = "0.5.0", optional = true }
 tinytools = { path = "../../vendor/tinytools/crates/tinytools", version = "0.6.0" }
 # Shared URL admission for opt-in remote multimodal attachments.
 tinytools-std = { path = "../../vendor/tinytools/crates/tinytools-std", version = "0.6.0", optional = true }

crates/tinyagents-harness/src/agent_loop/mod_tests.rs

@@ -9121,6 +9121,64 @@ async fn request_reasoning_wins_over_the_run_policy_default() {
     );
 }
 
+#[tokio::test]
+async fn thread_reasoning_snapshot_flows_through_policy_and_request_override_wins() {
+    use crate::testkit::ScriptedModel;
+
+    let store = crate::thread_reasoning::ThreadReasoningStore::default();
+    let thread = crate::ids::ThreadId::new("conversation-a");
+    store.set_effort(&thread, ReasoningEffort::High);
+    let turn_snapshot = store.reasoning_for_turn(Some(&thread), Some(4_000));
+    store.set_effort(&thread, ReasoningEffort::Low);
+
+    let model = Arc::new(ScriptedModel::replies(vec!["first", "second", "third"]));
+    let mut harness: AgentHarness<()> = AgentHarness::new();
+    harness.register_model("mock", model.clone());
+    harness.with_policy(RunPolicy {
+        default_reasoning: turn_snapshot,
+        ..RunPolicy::default()
+    });
+
+    harness
+        .invoke_default(&(), vec![Message::user("first turn")])
+        .await
+        .expect("first run succeeds");
+
+    let requests = model.requests();
+    let first = requests.first().expect("first model call");
+    let reasoning = first.reasoning.as_ref().expect("reasoning configured");
+    assert_eq!(reasoning.effort, Some(ReasoningEffort::High));
+    assert_eq!(reasoning.budget_tokens, Some(2_200));
+
+    harness.with_policy(RunPolicy {
+        default_reasoning: store.reasoning_for_turn(Some(&thread), Some(4_000)),
+        ..RunPolicy::default()
+    });
+    harness
+        .invoke_default(&(), vec![Message::user("next turn")])
+        .await
+        .expect("next run succeeds");
+    let requests = model.requests();
+    let next_reasoning = requests[1]
+        .reasoning
+        .as_ref()
+        .expect("reasoning configured");
+    assert_eq!(next_reasoning.effort, Some(ReasoningEffort::Low));
+
+    harness.push_middleware(Arc::new(RequestReasoningEffort(ReasoningEffort::XHigh)));
+    harness
+        .invoke_default(&(), vec![Message::user("request override")])
+        .await
+        .expect("request override run succeeds");
+    let requests = model.requests();
+    let reasoning = requests[2]
+        .reasoning
+        .as_ref()
+        .expect("request reasoning configured");
+    assert_eq!(reasoning.effort, Some(ReasoningEffort::XHigh));
+    assert_eq!(reasoning.budget_tokens, None);
+}
+
 // ── wave 1 follow-ups: batch termination and mixed truncated turns ───────────
 
 #[tokio::test]

crates/tinyagents-harness/src/cost/mod_tests.rs

@@ -214,6 +214,7 @@ fn estimates_each_component() {
         cache_creation_tokens: 200,
         reasoning_tokens: 10,
         charged_amount: None,
+        charged_amount_nanos: None,
         context_window_tokens: None,
     };
     let cost = estimate_cost(&pricing(), &usage);
@@ -243,6 +244,7 @@ fn cache_and_reasoning_tokens_are_not_double_counted() {
         cache_creation_tokens: 0,
         reasoning_tokens: 0,
         charged_amount: None,
+        charged_amount_nanos: None,
         context_window_tokens: None,
     };
     let cost = estimate_cost(&pricing(), &usage);
@@ -328,6 +330,7 @@ fn tiered_pricing_falls_back_to_flat_rate_for_unset_tier_fields() {
         cache_creation_tokens: 0,
         reasoning_tokens: 0,
         charged_amount: None,
+        charged_amount_nanos: None,
         context_window_tokens: None,
     };
     let cost = estimate_cost(&pricing, &usage);

crates/tinyagents-harness/src/events/mod.rs

@@ -38,6 +38,7 @@ pub mod ordered_sender;
 pub use call_dedupe::{CallDedupe, IterationCursor, advance_iteration};
 pub use cap_pauser::{CapPauser, PauseRecorder};
 pub use compaction_log::log_compacted;
+pub(crate) use generation_enrichment::is_valid_provider_id;
 pub use generation_enrichment::{GenerationEnrichment, GenerationPricing};
 pub use ordered_sender::{DEFAULT_MAX_OVERFLOW, OrderedSender, SendFailure};
 pub use types::*;

crates/tinyagents-harness/src/host/README.md

@@ -50,6 +50,7 @@ embedding, not as a way to fake "unconfigured."
 | `ContextComposer`, `StaticContextComposer`, `TurnContextRequest` | `context_composer` | Builds a turn's system prompt and preamble (required). |
 | `SecurityGate`, `AllowAllSecurityGate`, `GateDecision`, `ScreenOutcome`, `ToolCallRequest`, `ContentOrigin` | `security_gate` | Authorizes tool calls, screens untrusted text (required). |
 | `ModelResolver<State>`, `FixedModelResolver`, `PrebuiltModelResolver`, `ModelResolveRequest` | `model_resolver` | Chooses the model for a turn (required). |
+| `ModelFactory<State>`, `ModelBuildGate`, `ModelBuildRequest`, `BuildOpts`, `ModelPurpose`, `TurnModelSet` | `model_factory` | Builds role-keyed primary, workload, and summarization models with host policy checks. See [`model_factory/README.md`](model_factory/README.md) for assembly and provider boundaries. |
 | `AgentMemory`, `InMemoryAgentMemory`, `MemoryId`, `MemoryItem`, `NewMemory`, `RecallRequest` | `agent_memory` | Durable user memory (optional). |
 | `BudgetGate`, `UnlimitedBudgetGate`, `Permit`, `CallEstimate`, `ContextState`, `CompressionHint` | `budget_gate` | Admission, spend accounting, compression advice (optional). |
 | `ProgressSink`, `NoopProgressSink`, `RecordingProgressSink`, `ProgressEvent` | `progress_sink` | Coarse turn progress for a UI (optional). |
@@ -67,6 +68,7 @@ embedding, not as a way to fake "unconfigured."
 | `context_composer.rs` | `ContextComposer` trait, `TurnContextRequest`, `StaticContextComposer`. |
 | `experience_store.rs` | `ExperienceStore` trait, `Experience`, `InMemoryExperienceStore`. |
 | `learning_sink.rs` | `LearningSink` trait, `TurnSummary`, `NoopLearningSink`. |
+| `model_factory.rs` and `model_factory/` | Thin re-export plus the model-factory module; see its [module README](model_factory/README.md) for file ownership and operational constraints. |
 | `model_resolver.rs` | `ModelResolver<State>` trait, `ModelResolveRequest`, `FixedModelResolver`, and `PrebuiltModelResolver` for lead/pin/hint routes. |
 | `progress_sink.rs` | `ProgressEvent`, `ProgressSink` trait, `NoopProgressSink`, `RecordingProgressSink`. |
 | `security_gate.rs` | `SecurityGate` trait, `ToolCallRequest`, `GateDecision`, `ScreenOutcome`, `ContentOrigin`. |
@@ -112,6 +114,14 @@ composed transcript, and a security-authorized tool set, which is what keeps
 the loop reusable for both hosted and unhosted (explicit-model) callers.
 
 ModelResolveRequest carries model_hint separately from model_pin. Hints are
-host-interpreted routing vocabulary; pins are concrete model identifiers.
-Definition hints reach every hosted provider resolution, unless an exact model
-override is present. Missing hints in older serialized requests remain None.
+host-interpreted routing vocabulary. A construction-time model pin may be a
+requested alias; `ResolvedModelRoute` records the concrete provider and model
+the host selected.
+
+`ModelBuildRequest` is a construction-time request. Hosts gate each request
+before factory construction, then can use `TurnModelSet::map_models` to apply a
+per-turn wrapper to every built model before obtaining an owned resolver,
+ordered registrations, or summarizer. Models and resolvers obtained before
+mapping retain the earlier model instances. Definition hints reach every
+hosted provider resolution, unless an exact model override is present. Missing
+hints in older serialized requests remain None.

crates/tinyagents-harness/src/host/mod.rs

@@ -40,6 +40,7 @@ pub mod budget_gate;
 pub mod context_composer;
 pub mod experience_store;
 pub mod learning_sink;
+pub mod model_factory;
 pub mod model_resolver;
 pub mod progress_sink;
 pub mod security_gate;
@@ -56,6 +57,10 @@ pub use experience_store::{Experience, ExperienceStore, InMemoryExperienceStore}
 pub use learning_sink::{
     LearningSink, NoopLearningSink, ToolOutcomeRecord, TurnSummary, classify_tool_failure,
 };
+pub use model_factory::{
+    BuildOpts, ModelBuildGate, ModelBuildRequest, ModelFactory, ModelPurpose, TurnModelSet,
+    TurnModelSetBuilder,
+};
 pub use model_resolver::{
     FixedModelResolver, ModelResolveRequest, ModelResolver, PrebuiltModelResolver,
 };

crates/tinyagents-harness/src/host/model_factory.rs

@@ -0,0 +1,16 @@
+//! Host-facing model construction API, split into request types, turn assembly,
+//! and provider-specific subprocess adapters.
+
+mod assembly;
+mod types;
+
+#[cfg(feature = "claude-code")]
+mod subprocess;
+
+pub use assembly::{TurnModelSet, TurnModelSetBuilder};
+pub use types::{BuildOpts, ModelBuildGate, ModelBuildRequest, ModelFactory, ModelPurpose};
+
+#[cfg(feature = "claude-code")]
+pub use subprocess::{
+    SubprocessProviderContext, build_claude_subprocess_model, build_claude_subprocess_model_with,
+};

crates/tinyagents-harness/src/host/model_factory/README.md

@@ -0,0 +1,36 @@
+# Host model factory
+
+This module assembles a turn's primary model, optional workload-tier models,
+and summarizer from host-owned construction policy. It keeps provider
+selection and authorization at the host boundary while giving the harness a
+stable description of each requested route.
+
+## Public surface
+
+`types.rs` defines the serde-compatible `BuildOpts`, `ModelPurpose`, and
+`ModelBuildRequest` values, plus the `ModelBuildGate` and `ModelFactory`
+contracts. Hosts implement the gate to authorize each construction request and
+the factory to return a model with its resolved provider/model route.
+
+`assembly.rs` owns `TurnModelSetBuilder` and `TurnModelSet`. It builds the
+primary route, independently builds workload tiers (omitting unavailable or
+gated tiers), builds the summarizer, and records the concrete routes and error
+slots for later resolution. A primary pin does not implicitly pin workload
+routes. An explicit summarization role with no summary pin uses that role's
+configured default.
+
+`subprocess.rs` adapts Claude Code and Claude Agent SDK route values to the
+existing provider implementations. Hosts must run `ModelBuildGate` before
+constructing these subprocess-backed models. The injected constructor supports
+tests and hosts that own credential or process setup themselves.
+
+## Operational constraints
+
+- This module describes construction requests; it does not own product
+  configuration, credentials, or provider policy.
+- Workload failures are omitted from the turn model set; primary and
+  summarization construction failures are returned to the host.
+- Route aliases and concrete provider/model identities remain distinct. Use
+  `ResolvedModelRoute` for the concrete result.
+- Keep value types limited to serde and standard-library data so hosts can
+  persist or transport requests without importing runtime-specific types.