blazen 0.1.153 → 0.1.154

package/index.d.ts

@@ -2612,7 +2612,7 @@ export type JsModelCache = ModelCache
  *
  * ```javascript
  * const manager = new ModelManager({ budgetGb: 8.0 });
- * await manager.register("llama-7b", model, 4_000_000_000);
+ * await manager.register("llama-7b", model, 4_000_000_000n);  // BigInt
  * await manager.load("llama-7b");
  * ```
  */
@@ -2633,7 +2633,7 @@ export declare class ModelManager {
    * Only local in-process providers (mistral.rs, llama.cpp, candle) can be
    * registered -- remote HTTP providers will throw.
    */
-  register(id: string, model: JsCompletionModel, vramEstimateBytes?: number | undefined | null): Promise<void>
+  register(id: string, model: JsCompletionModel, vramEstimateBytes?: bigint | undefined | null): Promise<void>
   /**
    * Load a model, evicting LRU models if the budget would be exceeded.
    *
@@ -2657,9 +2657,9 @@ export declare class ModelManager {
    */
   ensureLoaded(id: string): Promise<void>
   /** Total VRAM currently used by loaded models (in bytes). */
-  usedBytes(): Promise<number>
+  usedBytes(): Promise<bigint>
   /** Available VRAM within the budget (in bytes). */
-  availableBytes(): Promise<number>
+  availableBytes(): Promise<bigint>
   /** Status of all registered models. */
   status(): Promise<Array<JsModelStatus>>
 }
@@ -2719,6 +2719,21 @@ export declare class MusicProvider {
 }
 export type JsMusicProvider = MusicProvider
 
+/**
+ * A no-op emitter that drops every event.
+ *
+ * Useful as a default when no downstream observer is wired up:
+ *
+ * ```javascript
+ * const model = new UsageRecordingCompletionModel(base, new NoopUsageEmitter(), "openai");
+ * ```
+ */
+export declare class NoopUsageEmitter {
+  /** Construct a no-op emitter. */
+  constructor()
+}
+export type JsNoopUsageEmitter = NoopUsageEmitter
+
 /**
  * An `OpenAI`-compatible embedding model (Together, Cohere, Fireworks).
  *
@@ -2873,6 +2888,37 @@ export declare class ParallelStage {
 }
 export type JsParallelStage = ParallelStage
 
+/**
+ * Fan out into multiple parallel sub-workflow branches.
+ *
+ * Each branch is a `SubWorkflowStep` that runs concurrently. The
+ * `joinStrategy` controls whether the parent waits for all branches
+ * (`JoinStrategy.WaitAll`) or only the first to complete
+ * (`JoinStrategy.FirstCompletes`).
+ */
+export declare class ParallelSubWorkflowsStep {
+  /**
+   * Create a parallel sub-workflow fan-out step.
+   *
+   * `branches` is an array of already-constructed `SubWorkflowStep`
+   * instances. The branches' inner workflows are captured by reference
+   * so the parent step keeps a stable view even if the originals are
+   * dropped from JS.
+   *
+   * `joinStrategy` defaults to `JoinStrategy.WaitAll`.
+   */
+  constructor(name: string, accepts: Array<string>, emits: Array<string>, branches: Array<SubWorkflowStep>, joinStrategy?: JoinStrategy | undefined | null)
+  /** The step name. */
+  get name(): string
+  /** Event type identifiers this step accepts. */
+  get accepts(): Array<string>
+  /** Event type identifiers this step may emit. */
+  get emits(): Array<string>
+  /** The join strategy used to combine branch results. */
+  get joinStrategy(): JoinStrategy
+}
+export type JsParallelSubWorkflowsStep = ParallelSubWorkflowsStep
+
 /** A Perplexity chat completion provider. */
 export declare class PerplexityProvider {
   /** Create a new Perplexity provider. */
@@ -2897,6 +2943,12 @@ export declare class Pipeline {
    * Consumes the pipeline -- calling start/resume a second time errors.
    */
   start(input: any): Promise<PipelineHandler>
+  /**
+   * Inspect the pipeline-level default retry configuration, if any.
+   * Mirrors [`blazen_pipeline::Pipeline::retry_config`] (Wave 2).
+   * Returns `null` after the pipeline has been consumed.
+   */
+  retryConfig(): JsRetryConfig | null
   /**
    * Resume the pipeline from a previously captured snapshot.
    * Consumes the pipeline.
@@ -2914,6 +2966,31 @@ export declare class PipelineBuilder {
   parallel(parallel: JsParallelStage): this
   /** Set a per-stage timeout in seconds. Each stage's workflow gets this duration. */
   timeoutPerStage(seconds: number): this
+  /**
+   * Set a total wall-clock timeout for the entire pipeline run, in
+   * seconds. When the pipeline does not finish within this duration the
+   * run is cancelled with a `PipelineError::Timeout`. Default is no
+   * total timeout. Mirrors [`blazen_pipeline::PipelineBuilder::total_timeout`].
+   */
+  totalTimeout(seconds: number): this
+  /** Disable the total-pipeline timeout (default). */
+  noTotalTimeout(): this
+  /**
+   * Set a default retry configuration applied to every LLM call inside
+   * the pipeline. Workflow / step / per-call overrides take precedence.
+   * Mirrors [`blazen_pipeline::PipelineBuilder::retry_config`].
+   */
+  retryConfig(config: JsRetryConfig): this
+  /**
+   * Disable pipeline-level retries (`maxRetries = 0`). Workflow / step /
+   * per-call overrides still take precedence.
+   */
+  noRetry(): this
+  /**
+   * Clear any pipeline-level retry config; LLM calls fall through to
+   * workflow / step / provider defaults.
+   */
+  clearRetryConfig(): this
   /**
    * Register a persist callback that receives a typed `PipelineSnapshot`
    * after each stage completes.
@@ -2954,6 +3031,13 @@ export declare class PipelineHandler {
   resumeInPlace(): Promise<void>
   /** Capture a snapshot without stopping the pipeline. */
   snapshot(): Promise<JsPipelineSnapshot>
+  /**
+   * Best-effort polled view of the pipeline's stage cursor. Mirrors
+   * [`blazen_pipeline::PipelineHandler::progress`].
+   *
+   * Returns `null` after [`Self::result`] has consumed the handler.
+   */
+  progress(): Promise<JsProgressSnapshot | null>
   /** Abort the pipeline. */
   abort(): Promise<void>
   /**
@@ -2971,6 +3055,16 @@ export declare class PipelineResult {
   get finalOutput(): any
   get stageResults(): Array<StageResult>
   get sharedState(): any
+  /**
+   * Aggregated token usage across the pipeline run. Mirrors
+   * [`blazen_pipeline::PipelineResult::usage_total`] (Wave 3).
+   */
+  get usageTotal(): JsTokenUsageClass
+  /**
+   * Aggregated cost in USD across the pipeline run. Mirrors
+   * [`blazen_pipeline::PipelineResult::cost_total_usd`] (Wave 3).
+   */
+  get costTotalUsd(): number
 }
 export type JsPipelineResult = PipelineResult
 
@@ -3337,6 +3431,34 @@ export declare class RetryCompletionModel {
 }
 export type JsRetryCompletionModel = RetryCompletionModel
 
+/**
+ * A `MemoryBackend` decorator that retries transient errors with
+ * exponential backoff.
+ *
+ * Mirrors `RetryCompletionModel` for `MemoryBackend`. Use one of the
+ * `wrapInMemory` / `wrapJsonl` / `wrapValkey` factories to wrap the
+ * matching backend.
+ *
+ * ```javascript
+ * const inner = new InMemoryBackend();
+ * const retried = RetryMemoryBackend.wrapInMemory(inner, { maxRetries: 5 });
+ * ```
+ */
+export declare class RetryMemoryBackend {
+  /** Wrap an `InMemoryBackend` with retry-on-transient-error behaviour. */
+  static wrapInMemory(backend: InMemoryBackend, config?: JsRetryConfig | undefined | null): RetryMemoryBackend
+  /** Wrap a `JsonlBackend` with retry-on-transient-error behaviour. */
+  static wrapJsonl(backend: JsonlBackend, config?: JsRetryConfig | undefined | null): RetryMemoryBackend
+  /** Wrap a `ValkeyBackend` with retry-on-transient-error behaviour. */
+  static wrapValkey(backend: ValkeyBackend, config?: JsRetryConfig | undefined | null): RetryMemoryBackend
+  /**
+   * Generic factory accepting any of the three concrete backends. Useful
+   * when the caller doesn't statically know which backend is in hand.
+   */
+  static wrap(backend: AnyBackend, config?: JsRetryConfig | undefined | null): RetryMemoryBackend
+}
+export type JsRetryMemoryBackend = RetryMemoryBackend
+
 /**
  * Built-in middleware that wraps the inner model with retry-on-transient-
  * error behaviour. Equivalent to constructing a
@@ -3525,6 +3647,17 @@ export declare class StageResult {
   get output(): any
   get skipped(): boolean
   get durationMs(): number
+  /**
+   * Token usage for this stage, if any LLM calls inside the stage
+   * emitted [`UsageEvent`](blazen_events::UsageEvent)s. Mirrors
+   * [`blazen_pipeline::StageResult::usage`] (Wave 3).
+   */
+  get usage(): JsTokenUsageClass | null
+  /**
+   * Cost in USD for this stage, if known. Mirrors
+   * [`blazen_pipeline::StageResult::cost_usd`] (Wave 3).
+   */
+  get costUsd(): number | null
 }
 export type JsStageResult = StageResult
 
@@ -3774,6 +3907,41 @@ export declare class StructuredOutput {
 }
 export type JsStructuredOutput = StructuredOutput
 
+/**
+ * A workflow step that delegates to another `Workflow`.
+ *
+ * The parent workflow's event loop spawns the child via `Workflow.run()`,
+ * converts the parent event to JSON for the child's input, and wraps the
+ * child's terminal `StopEvent.result` into a `DynamicEvent` named
+ * `"<stepName>::output"` for the parent.
+ *
+ * ```javascript
+ * const child = new Workflow("enrich");
+ * child.addStep("enrich", ["blazen::StartEvent"], async (ev) => ({ type: "blazen::StopEvent", result: { ok: true } }));
+ * const step = new SubWorkflowStep("enrich", ["blazen::StartEvent"], ["enrich::output"], child);
+ * const parent = new Workflow("parent");
+ * parent.addSubworkflowStepObj(step);
+ * ```
+ */
+export declare class SubWorkflowStep {
+  /**
+   * Create a sub-workflow step.
+   *
+   * `name` / `accepts` / `emits` describe routing. `inner` is the child
+   * workflow whose event loop is spawned for each parent dispatch. The
+   * inner workflow is cloned (and built) at construction time so this
+   * step instance can be reused across builders.
+   */
+  constructor(name: string, accepts: Array<string>, emits: Array<string>, inner: JsWorkflow, timeoutSecs?: number | undefined | null, retryConfig?: JsRetryConfig | undefined | null)
+  /** The step name. */
+  get name(): string
+  /** Event type identifiers this step accepts. */
+  get accepts(): Array<string>
+  /** Event type identifiers this step may emit. */
+  get emits(): Array<string>
+}
+export type JsSubWorkflowStep = SubWorkflowStep
+
 /**
  * r" Base class for 3D model generation providers.
  * r"
@@ -4113,9 +4281,93 @@ export declare class TypedTool {
   get description(): string
   /** The JSON Schema parameter definition. */
   get parameters(): any
+  /**
+   * Whether this tool is marked as an "exit" tool, signalling the agent
+   * loop to stop after it runs. Mirrors
+   * [`blazen_llm::Tool::is_exit`] (Wave 6).
+   */
+  get isExit(): boolean
+  /**
+   * Builder-style toggle for the exit-tool marker. Returns a new
+   * [`JsTypedTool`] that shares the same underlying handler but
+   * reports the requested [`Tool::is_exit`] value. Mirrors
+   * [`blazen_llm::TypedTool::exit_tool`] (Wave 6).
+   *
+   * ```typescript
+   * const exit = new TypedTool(
+   *   "submit",
+   *   "Submit the final answer and exit",
+   *   { type: "object", properties: { answer: { type: "string" } }, required: ["answer"] },
+   *   async (_n, args) => ({ submitted: args.answer }),
+   * ).exitTool(true);
+   * ```
+   */
+  exitTool(exit: boolean): TypedTool
 }
 export type JsTypedTool = TypedTool
 
+/**
+ * A sink for emitted [`JsUsageEvent`]s.
+ *
+ * Construct with a JS callback that handles each event. The callback runs
+ * on the libuv main thread, so it can do anything synchronous; a thrown
+ * error is caught and logged, never propagated into the completion call.
+ *
+ * ```javascript
+ * const events: UsageEvent[] = [];
+ * const emitter = new UsageEmitter((event) => { events.push(event); });
+ * const model = new UsageRecordingCompletionModel(base, emitter, "openai");
+ * ```
+ */
+export declare class UsageEmitter {
+  /**
+   * Create an emitter from a JS callback. The callback is invoked once
+   * per emitted event.
+   */
+  constructor(callback: ((arg: UsageEvent) => void))
+}
+export type JsUsageEmitter = UsageEmitter
+
+/**
+ * A `CompletionModel` decorator that emits a `UsageEvent` after each
+ * successful `complete` call. Mirrors
+ * `blazen_llm::usage_recording::UsageRecordingCompletionModel`.
+ *
+ * ```javascript
+ * const base = CompletionModel.openai();
+ * const events = [];
+ * const emitter = new UsageEmitter((e) => events.push(e));
+ * const model = new UsageRecordingCompletionModel(base, emitter, "openai");
+ * const response = await model.complete([ChatMessage.user("hi")]);
+ * ```
+ */
+export declare class UsageRecordingCompletionModel {
+  /** Wrap a `CompletionModel` with a usage-recording layer. */
+  constructor(model: CompletionModel, emitter: AnyEmitter, providerLabel: string, runId?: string | undefined | null)
+  /** The underlying provider's model id. */
+  get modelId(): string
+  /**
+   * Convert this decorator into a `CompletionModel` so it can be passed to
+   * APIs that expect the base type (`runAgent`, further decorators, …).
+   */
+  toCompletionModel(): CompletionModel
+}
+export type JsUsageRecordingCompletionModel = UsageRecordingCompletionModel
+
+/**
+ * An `EmbeddingModel` decorator that emits a `UsageEvent` after each
+ * successful `embed` call.
+ */
+export declare class UsageRecordingEmbeddingModel {
+  /** Wrap an `EmbeddingModel` with a usage-recording layer. */
+  constructor(model: EmbeddingModel, emitter: AnyEmitter, providerLabel: string, runId?: string | undefined | null)
+  /** The underlying provider's model id. */
+  get modelId(): string
+  /** Output dimensionality. */
+  get dimensions(): number
+}
+export type JsUsageRecordingEmbeddingModel = UsageRecordingEmbeddingModel
+
 /**
  * A Valkey/Redis-backed backend for the memory store.
  *
@@ -4307,6 +4559,45 @@ export declare class Workflow {
    * started, step completed, step failed). Defaults to `false`.
    */
   setAutoPublishEvents(enabled: boolean): void
+  /**
+   * Register a sub-workflow step that delegates to another `Workflow`.
+   * Mirrors [`blazen_core::WorkflowBuilder::add_subworkflow_step`].
+   *
+   * - `name`: human-readable step name.
+   * - `accepts`: array of event type strings this step handles.
+   * - `emits`: array of event type strings this step may emit (informational).
+   * - `inner`: the child `Workflow` to run.
+   * - `timeoutSecs`: optional wall-clock timeout for the child run.
+   * - `retryConfig`: optional retry policy for the child run.
+   */
+  addSubworkflowStep(name: string, accepts: Array<string>, emits: Array<string>, inner: Workflow, timeoutSecs?: number | undefined | null, retryConfig?: JsRetryConfig | undefined | null): void
+  /**
+   * Register a parallel sub-workflows fan-out step. Mirrors
+   * [`blazen_core::WorkflowBuilder::add_parallel_subworkflows`].
+   *
+   * `branchSpecs` and `branchWorkflows` are parallel arrays of equal
+   * length: each `(spec, wf)` pair becomes one branch. They're split
+   * into two parameters because napi-rs cannot embed napi-class values
+   * (the `Workflow` instances) inside `#[napi(object)]` shapes.
+   *
+   * `joinStrategy` defaults to `WaitAll`.
+   */
+  addParallelSubworkflows(name: string, accepts: Array<string>, emits: Array<string>, branchSpecs: Array<SubWorkflowBranchSpec>, branchWorkflows: Array<Workflow>, joinStrategy?: JoinStrategy | undefined | null): void
+  /**
+   * Register a pre-built [`SubWorkflowStep`] wrapper.
+   *
+   * Object-form of [`Self::add_subworkflow_step`]: the same step instance
+   * can be reused across multiple workflows since its inner child
+   * workflow is captured in `Arc` form at construction time.
+   */
+  addSubworkflowStepObj(step: SubWorkflowStep): void
+  /**
+   * Register a pre-built [`ParallelSubWorkflowsStep`] wrapper.
+   *
+   * Object-form of [`Self::add_parallel_subworkflows`]: lifts the
+   * awkward parallel-arrays signature into a single class instance.
+   */
+  addParallelSubworkflowsObj(step: ParallelSubWorkflowsStep): void
   /**
    * Add a step to the workflow.
    *
@@ -4462,6 +4753,31 @@ export declare class WorkflowBuilder {
    * `SessionPausePolicy.PickleOrError`.
    */
   sessionPausePolicy(policy: SessionPausePolicy): this
+  /**
+   * Set a per-step wall-clock timeout in seconds on the most-recently
+   * added step. Mirrors [`blazen_core::WorkflowBuilder::step_timeout`].
+   * Errors if no step has been registered yet.
+   */
+  stepTimeout(seconds: number): this
+  /** Clear the per-step timeout on the most-recently added step. */
+  noStepTimeout(): this
+  /**
+   * Set a per-step retry config on the most-recently added step.
+   * Mirrors [`blazen_core::WorkflowBuilder::step_retry`].
+   */
+  stepRetry(config: JsRetryConfig): this
+  /** Disable retries on the most-recently added step (`maxRetries = 0`). */
+  noStepRetry(): this
+  /**
+   * Set a workflow-level default retry config. Step / per-call
+   * overrides take precedence; pipeline / provider defaults take
+   * lower precedence.
+   */
+  retryConfig(config: JsRetryConfig): this
+  /** Disable workflow-level retries (`maxRetries = 0`). */
+  noRetry(): this
+  /** Clear any workflow-level retry config. */
+  clearRetryConfig(): this
   /**
    * Enable history collection.
    *
@@ -4611,6 +4927,19 @@ export declare class WorkflowHandler {
    * they may have built earlier.
    */
   respondToInputTyped(event: InputResponseEvent): Promise<void>
+  /**
+   * Aggregated token usage across the workflow run so far.
+   *
+   * Mirrors [`blazen_core::WorkflowHandler::usage_total`]. Returns
+   * `null` after the handler has been consumed by [`Self::result`].
+   */
+  usageTotal(): Promise<TokenUsage | null>
+  /**
+   * Aggregated cost in USD across the workflow run so far. Mirrors
+   * [`blazen_core::WorkflowHandler::cost_total_usd`]. Returns `null`
+   * after the handler has been consumed by [`Self::result`].
+   */
+  costTotalUsd(): Promise<number | null>
   /** Abort the running workflow. */
   abort(): Promise<void>
   /**
@@ -4731,6 +5060,19 @@ export declare class XaiProvider {
 }
 export type JsXaiProvider = XaiProvider
 
+/**
+ * Aggregate one [`JsUsageEvent`] into a [`crate::types::JsTokenUsageClass`].
+ * Returns a fresh class instance that adds the seven token counters from the
+ * event into the existing usage. Mirrors the Rust `TokenUsage::add` /
+ * `PipelineState::record_usage` flow at the JS layer.
+ *
+ * # Errors
+ *
+ * Currently never returns an error; the [`Result`] return is reserved for
+ * future validation (e.g. parsing the `runId` UUID).
+ */
+export declare function addUsageToTokenUsage(base: JsTokenUsageClass, event: UsageEvent): JsTokenUsageClass
+
 /**
  * Typed configuration for the agentic tool execution loop.
  *
@@ -4742,8 +5084,23 @@ export type JsXaiProvider = XaiProvider
 export interface AgentConfig {
   /** Maximum number of tool call rounds before forcing a stop. */
   maxIterations: number
-  /** Whether to add an implicit "finish" tool the model can call to exit early. */
+  /**
+   * Whether to add the legacy implicit "finish" tool the model can call
+   * to exit early.
+   */
   addFinishTool: boolean
+  /**
+   * Suppress automatic registration of the built-in `finish_workflow`
+   * exit tool. Default `false`. Mirrors
+   * [`blazen_llm::AgentConfig::no_finish_tool`] (Wave 6).
+   */
+  noFinishTool: boolean
+  /**
+   * Override the name of the built-in `finish_workflow` tool. `null`
+   * uses the canonical [`crate::agent::FINISH_WORKFLOW_TOOL_NAME`].
+   * Mirrors [`blazen_llm::AgentConfig::finish_tool_name`] (Wave 6).
+   */
+  finishToolName?: string
   /** Optional system prompt prepended to messages. */
   systemPrompt?: string
   /** Sampling temperature. */
@@ -4948,6 +5305,14 @@ export interface CompletionRequest {
   audioConfig?: any
 }
 
+/**
+ * Compute the cost in USD for an audio call (TTS / STT) given the model id
+ * and the duration in seconds. Returns `null` when the model has no
+ * `perSecond` pricing entry registered. Mirrors
+ * [`blazen_llm::compute_audio_cost`].
+ */
+export declare function computeAudioCost(modelId: string, seconds: number): number | null
+
 /**
  * Compute ELID-based similarity between two embedding vectors.
  *
@@ -4965,6 +5330,18 @@ export declare function computeElidSimilarity(a: Array<number>, b: Array<number>
  */
 export declare function computeEmbeddingSimhashSimilarity(a: Array<number>, b: Array<number>): number
 
+/**
+ * Compute the cost in USD for an image-generation call given the model id
+ * and the number of images returned. Returns `null` when the model has no
+ * `perImage` pricing entry registered. Mirrors
+ * [`blazen_llm::compute_image_cost`].
+ *
+ * ```javascript
+ * const cost = computeImageCost("dall-e-3", 4);
+ * ```
+ */
+export declare function computeImageCost(modelId: string, imageCount: number): number | null
+
 /**
  * Compute similarity between two text strings using 64-bit `SimHash`.
  *
@@ -4973,6 +5350,14 @@ export declare function computeEmbeddingSimhashSimilarity(a: Array<number>, b: A
  */
 export declare function computeTextSimhashSimilarity(a: string, b: string): number
 
+/**
+ * Compute the cost in USD for a video-generation call given the model id
+ * and the output duration in seconds. Returns `null` when the model has no
+ * `perSecond` pricing entry registered. Mirrors
+ * [`blazen_llm::compute_video_cost`].
+ */
+export declare function computeVideoCost(modelId: string, seconds: number): number | null
+
 /**
  * Estimate the total token count for an array of chat messages.
  *
@@ -4999,6 +5384,9 @@ export interface CustomProviderOptions {
   providerId?: string
 }
 
+/** Build a default [`JsHttpClientConfig`] (60s request, 10s connect, no UA). */
+export declare function defaultHttpClientConfig(): HttpClientConfig
+
 /**
  * Configuration for subclassed `EmbeddingModel` instances.
  *
@@ -5097,6 +5485,30 @@ export interface FileContent {
   filename?: string
 }
 
+/**
+ * Canonical name of the built-in `finish_workflow` exit tool. Mirrors
+ * [`blazen_llm::FINISH_WORKFLOW_TOOL_NAME`].
+ */
+export const FINISH_WORKFLOW_TOOL_NAME: string
+
+/**
+ * Alias of [`finish_workflow_tool_def`] under the canonical Rust name
+ * (`finish_workflow_tool` → `finishWorkflowTool` in JS). Returns a
+ * `JsToolDef` for the built-in exit tool.
+ *
+ * Mirrors `blazen_llm::finish_workflow_tool` and `blazen-py`'s
+ * `finish_workflow_tool()` for cross-binding parity.
+ */
+export declare function finishWorkflowTool(): JsToolDef
+
+/**
+ * Build a fresh JSON-Schema description of the built-in `finish_workflow`
+ * exit tool. The shape mirrors [`blazen_llm::finish_workflow_tool`] —
+ * callers that want to surface the same tool to a JS-side agent loop can
+ * use this object as a [`JsToolDef`] entry.
+ */
+export declare function finishWorkflowToolDef(): JsToolDef
+
 /**
  * Format the tail of a `ProviderHttp` error message, the same way
  * `BlazenError::ProviderHttp` does internally.
@@ -5153,6 +5565,40 @@ export declare const enum HistoryEventKindTag {
   WorkflowTimedOut = 'WorkflowTimedOut'
 }
 
+/**
+ * Configuration applied when constructing the default HTTP client.
+ *
+ * Mirrors [`HttpClientConfig`]. All fields are optional in JS — pass
+ * `null` / `undefined` for any field to mean "no timeout / no UA
+ * override".
+ *
+ * ```typescript
+ * const cfg: HttpClientConfig = {
+ *   requestTimeoutMs: 30_000,
+ *   connectTimeoutMs: 5_000,
+ *   userAgent: "my-app/1.0",
+ * };
+ * const unlimited = HttpClientConfig.unlimited();
+ * ```
+ */
+export interface HttpClientConfig {
+  /**
+   * Maximum wall-clock duration for a single request, in milliseconds.
+   * `null` / `undefined` means unlimited.
+   */
+  requestTimeoutMs?: number
+  /**
+   * Maximum duration for the connection-establishment phase, in
+   * milliseconds. `null` / `undefined` means unlimited.
+   */
+  connectTimeoutMs?: number
+  /**
+   * User-Agent header string. `null` / `undefined` uses the underlying
+   * client's default.
+   */
+  userAgent?: string
+}
+
 /**
  * Initialize the Langfuse exporter and install it as a layer on the global
  * `tracing` subscriber.
@@ -5252,10 +5698,22 @@ export interface JsAgentRunOptions {
   /** Maximum tokens per completion call. */
   maxTokens?: number
   /**
-   * Whether to add a built-in "finish" tool that the model can call to
-   * signal it has a final answer.
+   * Whether to add the legacy implicit "finish" tool the model can call
+   * to signal it has a final answer.
    */
   addFinishTool?: boolean
+  /**
+   * Suppress automatic registration of the built-in `finish_workflow`
+   * exit tool (default `false`, i.e. it is auto-added). Mirrors
+   * [`blazen_llm::AgentConfig::no_finish_tool`] (Wave 6).
+   */
+  noFinishTool?: boolean
+  /**
+   * Override the name of the built-in `finish_workflow` tool. Defaults
+   * to [`FINISH_WORKFLOW_TOOL_NAME`]. Mirrors
+   * [`blazen_llm::AgentConfig::finish_tool_name`] (Wave 6).
+   */
+  finishToolName?: string
   /**
    * Maximum number of tool calls to execute concurrently within a single
    * model response. `0` means unlimited (all in parallel). Defaults to 0.
@@ -5319,6 +5777,8 @@ export interface JsAudioResult {
   audio: Array<JsGeneratedAudio>
   timing: JsRequestTiming
   cost?: number
+  usage?: JsTokenUsage
+  audioSeconds?: number
   metadata: any
 }
 
@@ -5768,6 +6228,8 @@ export interface JsImageResult {
   images: Array<JsGeneratedImage>
   timing: JsRequestTiming
   cost?: number
+  usage?: JsTokenUsage
+  imageCount?: number
   metadata: any
 }
 
@@ -5922,7 +6384,7 @@ export interface JsModelStatus {
   /** Whether the model is currently loaded into VRAM. */
   loaded: boolean
   /** Estimated VRAM footprint in bytes. */
-  vramEstimate: number
+  vramEstimate: bigint
 }
 
 export interface JsMusicRequest {
@@ -6239,6 +6701,7 @@ export interface JsThreeDResult {
   models: Array<JsGenerated3DModel>
   timing: JsRequestTiming
   cost?: number
+  usage?: JsTokenUsage
   metadata: any
 }
 
@@ -6337,6 +6800,8 @@ export interface JsTranscriptionResult {
   language?: string
   timing: JsRequestTiming
   cost?: number
+  usage?: JsTokenUsage
+  audioSeconds?: number
   metadata: any
 }
 
@@ -6376,6 +6841,8 @@ export interface JsVideoResult {
   videos: Array<JsGeneratedVideo>
   timing: JsRequestTiming
   cost?: number
+  usage?: JsTokenUsage
+  videoSeconds?: number
   metadata: any
 }
 
@@ -6481,6 +6948,16 @@ export interface JsWorkflowResult {
   type: string
   /** The result data as a JSON object. */
   data: any
+  /**
+   * Aggregated token usage across the run, mirroring
+   * [`blazen_core::WorkflowResult::usage_total`].
+   */
+  usageTotal: JsTokenUsage
+  /**
+   * Aggregated cost in USD across the run, mirroring
+   * [`blazen_core::WorkflowResult::cost_total_usd`].
+   */
+  costTotalUsd: number
 }
 
 /**
@@ -6585,6 +7062,28 @@ export interface MessageContent {
   parts?: Array<JsContentPart>
 }
 
+/**
+ * Discriminant for the kind of provider call a [`JsUsageEvent`] describes.
+ *
+ * Mirrors [`blazen_events::Modality`]. The string-enum representation
+ * matches the `Modality::*` unit variants. Custom modalities (the Rust
+ * `Modality::Custom(String)` variant) are surfaced via the
+ * [`JsUsageEvent::modalityCustom`] string field — when `modalityCustom`
+ * is non-null, callers should treat `modality` as `Custom` regardless of
+ * its value.
+ */
+export declare const enum Modality {
+  Llm = 'Llm',
+  Embedding = 'Embedding',
+  ImageGen = 'ImageGen',
+  AudioTts = 'AudioTts',
+  AudioStt = 'AudioStt',
+  Video = 'Video',
+  ThreeD = 'ThreeD',
+  BackgroundRemoval = 'BackgroundRemoval',
+  Custom = 'Custom'
+}
+
 /**
  * Capability flags advertised by a model.
  *
@@ -6645,10 +7144,20 @@ export interface ModelInfo {
 export interface ModelManagerConfig {
   /** VRAM budget in gigabytes (e.g. `8.0` for 8 GiB). */
   budgetGb?: number
-  /** VRAM budget in bytes. */
-  budgetBytes?: number
+  /** VRAM budget in bytes (pass as JS `BigInt` to support values >4 GiB). */
+  budgetBytes?: bigint
 }
 
+/** Build an empty [`JsRetryStack`] with every scope set to `null`. */
+export declare function newRetryStack(): RetryStack
+
+/**
+ * Build a default [`JsUsageEvent`] for the given provider / model, with
+ * every numeric field zeroed and `modality = Llm`. Useful as a starting
+ * point for emitter shims that only know a subset of the fields.
+ */
+export declare function newUsageEvent(provider: string, model: string, runId: string): UsageEvent
+
 /**
  * Why a workflow was paused.
  *
@@ -6679,14 +7188,76 @@ export interface PersistedEvent {
  *
  * Mirrors [`blazen_llm::PricingEntry`]. The existing
  * [`crate::types::pricing::JsModelPricing`] is a richer "input" type that
- * also carries `perImage` / `perSecond`; this shape only covers the
- * per-million token rates the registry actually stores.
+ * also carries `perImage` / `perSecond`; this shape covers the
+ * per-million token rates the registry actually stores, plus optional
+ * per-image and per-second rates for multimodal/audio/video models.
  */
 export interface PricingEntry {
   /** USD per million input (prompt) tokens. */
   inputPerMillion: number
   /** USD per million output (completion) tokens. */
   outputPerMillion: number
+  /** USD per image (for multimodal models). `null` if not applicable. */
+  perImage?: number
+  /** USD per second (for audio/video models). `null` if not applicable. */
+  perSecond?: number
+}
+
+/**
+ * Per-stage / per-step progress tick emitted by Pipeline and Workflow
+ * runners. Mirrors [`blazen_events::ProgressEvent`].
+ *
+ * `total` and `percent` are absent (`undefined`) when the step set is
+ * dynamic and the total is not known up front.
+ */
+export interface ProgressEvent {
+  /** What this progress event describes. */
+  kind: ProgressKind
+  /** Current step / stage index (1-based). */
+  current: number
+  /** Total number of steps / stages, when known. */
+  total?: number
+  /** Progress as a percentage in `0.0..=100.0`, when computable. */
+  percent?: number
+  /** Human-readable label for this progress tick (typically the step name). */
+  label: string
+  /** UUID of the run this progress belongs to. */
+  runId: string
+}
+
+/**
+ * What a [`JsProgressEvent`] describes. Mirrors
+ * [`blazen_events::ProgressKind`].
+ */
+export declare const enum ProgressKind {
+  Pipeline = 'Pipeline',
+  Workflow = 'Workflow',
+  SubWorkflow = 'SubWorkflow',
+  Stage = 'Stage'
+}
+
+/**
+ * Lightweight, polled view of a running pipeline's progress.
+ *
+ * Mirrors [`ProgressSnapshot`]. Reads are best-effort and may briefly be
+ * one stage behind the actual position because they do not synchronise
+ * with the executor task.
+ */
+export interface ProgressSnapshot {
+  /**
+   * 1-based index of the stage currently executing (or just completed).
+   * `0` before the first stage starts.
+   */
+  currentStageIndex: number
+  /** Total number of stages declared on the pipeline. */
+  totalStages: number
+  /** Progress as a percentage in `0.0..=100.0`. */
+  percent: number
+  /**
+   * Name of the current stage, when available. Always `null` from the
+   * current atomic-index implementation; reserved for future use.
+   */
+  currentStageName?: string
 }
 
 /** Options for creating a `PromptTemplate`. */
@@ -6915,6 +7486,39 @@ export declare function resolveApiKey(provider: string, explicit?: string | unde
  */
 export declare function resolveBeerToken(): string | null
 
+/**
+ * Resolve the effective [`JsRetryConfig`] for the given stack and an
+ * optional per-call override. Mirrors [`RetryStack::resolve`].
+ *
+ * ```typescript
+ * const effective = resolveRetryStack(
+ *   { workflow: { maxRetries: 5 } },
+ *   { maxRetries: 9 }, // per-call override wins
+ * );
+ * // effective.maxRetries === 9
+ * ```
+ */
+export declare function resolveRetryStack(stack: RetryStack, callOverride?: JsRetryConfig | undefined | null): JsRetryConfig
+
+/**
+ * Snapshot of every scope's retry configuration. Mirrors
+ * [`RetryStack`].
+ *
+ * All fields are optional; any combination of `null` / `undefined`
+ * scopes is valid and falls through to the next-outer non-`None` scope
+ * when [`resolveRetryStack`] is called.
+ */
+export interface RetryStack {
+  /** Provider-level default (lowest priority). */
+  provider?: JsRetryConfig
+  /** Pipeline-level default. */
+  pipeline?: JsRetryConfig
+  /** Workflow-level override. */
+  workflow?: JsRetryConfig
+  /** Step-level override (highest priority before the per-call override). */
+  step?: JsRetryConfig
+}
+
 /**
  * Run an agentic tool execution loop.
  *
@@ -7145,6 +7749,26 @@ export interface StructuredResponse {
   artifacts: Array<JsArtifact>
 }
 
+/**
+ * Plain-object descriptor for one branch of a parallel fan-out — every
+ * field except `workflowName` is metadata. The actual child `Workflow`
+ * instances are passed alongside the spec array in
+ * [`JsWorkflow::addParallelSubworkflows`] (napi cannot embed napi-class
+ * values inside `#[napi(object)]` shapes).
+ */
+export interface SubWorkflowBranchSpec {
+  /** Human-readable name for this branch. */
+  name: string
+  /** Event types this branch's parent step accepts. */
+  accepts: Array<string>
+  /** Event types this branch's parent step may emit (informational). */
+  emits: Array<string>
+  /** Optional wall-clock timeout in seconds for this branch. */
+  timeoutSecs?: number
+  /** Optional retry config for this branch. */
+  retryConfig?: JsRetryConfig
+}
+
 /**
  * The role for a prompt template.
  *
@@ -7239,6 +7863,82 @@ export declare function tryDeserializeEvent(name: string, jsonStr: string): Prom
  */
 export declare function typedToolSimple(name: string, description: string, parameters: any, handler: TypedToolHandlerTsfn): TypedTool
 
+/**
+ * Build a [`JsHttpClientConfig`] with no request or connect timeout.
+ * Mirrors [`HttpClientConfig::unlimited`].
+ */
+export declare function unlimitedHttpClientConfig(): HttpClientConfig
+
+/**
+ * Token / cost / latency snapshot for a single provider call, emitted
+ * after each LLM / embedding / image / audio / video / 3D request.
+ *
+ * Pipelines and workflows aggregate these into [`PipelineState.usageTotal`]
+ * and [`PipelineState.costTotalUsd`] when a `UsageEmitter` is wired up.
+ *
+ * ```typescript
+ * import { UsageEvent, Modality } from 'blazen';
+ *
+ * const ev: UsageEvent = {
+ *     provider: "openai",
+ *     model: "gpt-4o-mini",
+ *     modality: Modality.Llm,
+ *     promptTokens: 100,
+ *     completionTokens: 25,
+ *     totalTokens: 125,
+ *     reasoningTokens: 0,
+ *     cachedInputTokens: 0,
+ *     audioInputTokens: 0,
+ *     audioOutputTokens: 0,
+ *     imageCount: 0,
+ *     audioSeconds: 0,
+ *     videoSeconds: 0,
+ *     latencyMs: 432,
+ *     costUsd: 0.000_25,
+ *     runId: "...",
+ * };
+ * ```
+ */
+export interface UsageEvent {
+  /** The provider that served the call (e.g. `"openai"`, `"anthropic"`). */
+  provider: string
+  /** The model identifier. */
+  model: string
+  /** Discriminant for the kind of call. */
+  modality: Modality
+  /**
+   * Free-form custom-modality label. Populated when [`Self::modality`] is
+   * [`JsModality::Custom`]; ignored otherwise.
+   */
+  modalityCustom?: string
+  /** Number of prompt / input tokens billed. */
+  promptTokens: number
+  /** Number of completion / output tokens billed. */
+  completionTokens: number
+  /** Total tokens billed (typically `promptTokens + completionTokens`). */
+  totalTokens: number
+  /** Reasoning tokens (e.g. `OpenAI` o-series, Anthropic extended thinking). */
+  reasoningTokens: number
+  /** Tokens served from the provider's prompt cache at a discount. */
+  cachedInputTokens: number
+  /** Audio input tokens (multimodal speech-in models). */
+  audioInputTokens: number
+  /** Audio output tokens (multimodal speech-out models). */
+  audioOutputTokens: number
+  /** Number of images generated or processed. */
+  imageCount: number
+  /** Audio duration in seconds (for STT inputs and TTS outputs). */
+  audioSeconds: number
+  /** Video duration in seconds. */
+  videoSeconds: number
+  /** Cost in USD as reported (or computed) for this call. */
+  costUsd?: number
+  /** Wall-clock latency of the provider call in milliseconds. */
+  latencyMs: number
+  /** UUID of the run / pipeline invocation this usage belongs to. */
+  runId: string
+}
+
 /** Returns the version of the blazen library. */
 export declare function version(): string
 

package/index.js

@@ -720,6 +720,8 @@ module.exports.ModelRegistry = nativeBinding.ModelRegistry
 module.exports.JsModelRegistry = nativeBinding.JsModelRegistry
 module.exports.MusicProvider = nativeBinding.MusicProvider
 module.exports.JsMusicProvider = nativeBinding.JsMusicProvider
+module.exports.NoopUsageEmitter = nativeBinding.NoopUsageEmitter
+module.exports.JsNoopUsageEmitter = nativeBinding.JsNoopUsageEmitter
 module.exports.OpenAiCompatEmbeddingModel = nativeBinding.OpenAiCompatEmbeddingModel
 module.exports.JsOpenAiCompatEmbeddingModel = nativeBinding.JsOpenAiCompatEmbeddingModel
 module.exports.OpenAiCompatProvider = nativeBinding.OpenAiCompatProvider
@@ -732,6 +734,8 @@ module.exports.OpenRouterProvider = nativeBinding.OpenRouterProvider
 module.exports.JsOpenRouterProvider = nativeBinding.JsOpenRouterProvider
 module.exports.ParallelStage = nativeBinding.ParallelStage
 module.exports.JsParallelStage = nativeBinding.JsParallelStage
+module.exports.ParallelSubWorkflowsStep = nativeBinding.ParallelSubWorkflowsStep
+module.exports.JsParallelSubWorkflowsStep = nativeBinding.JsParallelSubWorkflowsStep
 module.exports.PerplexityProvider = nativeBinding.PerplexityProvider
 module.exports.JsPerplexityProvider = nativeBinding.JsPerplexityProvider
 module.exports.Pipeline = nativeBinding.Pipeline
@@ -766,6 +770,8 @@ module.exports.RequestTiming = nativeBinding.RequestTiming
 module.exports.JsRequestTimingClass = nativeBinding.JsRequestTimingClass
 module.exports.RetryCompletionModel = nativeBinding.RetryCompletionModel
 module.exports.JsRetryCompletionModel = nativeBinding.JsRetryCompletionModel
+module.exports.RetryMemoryBackend = nativeBinding.RetryMemoryBackend
+module.exports.JsRetryMemoryBackend = nativeBinding.JsRetryMemoryBackend
 module.exports.RetryMiddleware = nativeBinding.RetryMiddleware
 module.exports.JsRetryMiddleware = nativeBinding.JsRetryMiddleware
 module.exports.SessionNamespace = nativeBinding.SessionNamespace
@@ -792,6 +798,8 @@ module.exports.StopEvent = nativeBinding.StopEvent
 module.exports.JsStopEventClass = nativeBinding.JsStopEventClass
 module.exports.StructuredOutput = nativeBinding.StructuredOutput
 module.exports.JsStructuredOutput = nativeBinding.JsStructuredOutput
+module.exports.SubWorkflowStep = nativeBinding.SubWorkflowStep
+module.exports.JsSubWorkflowStep = nativeBinding.JsSubWorkflowStep
 module.exports.ThreeDProvider = nativeBinding.ThreeDProvider
 module.exports.JsThreeDProvider = nativeBinding.JsThreeDProvider
 module.exports.TogetherProvider = nativeBinding.TogetherProvider
@@ -814,6 +822,12 @@ module.exports.TTSProvider = nativeBinding.TTSProvider
 module.exports.JsTTSProvider = nativeBinding.JsTTSProvider
 module.exports.TypedTool = nativeBinding.TypedTool
 module.exports.JsTypedTool = nativeBinding.JsTypedTool
+module.exports.UsageEmitter = nativeBinding.UsageEmitter
+module.exports.JsUsageEmitter = nativeBinding.JsUsageEmitter
+module.exports.UsageRecordingCompletionModel = nativeBinding.UsageRecordingCompletionModel
+module.exports.JsUsageRecordingCompletionModel = nativeBinding.JsUsageRecordingCompletionModel
+module.exports.UsageRecordingEmbeddingModel = nativeBinding.UsageRecordingEmbeddingModel
+module.exports.JsUsageRecordingEmbeddingModel = nativeBinding.JsUsageRecordingEmbeddingModel
 module.exports.ValkeyBackend = nativeBinding.ValkeyBackend
 module.exports.JsValkeyBackend = nativeBinding.JsValkeyBackend
 module.exports.ValkeyCheckpointStore = nativeBinding.ValkeyCheckpointStore
@@ -838,17 +852,25 @@ module.exports.WorkflowSnapshot = nativeBinding.WorkflowSnapshot
 module.exports.JsWorkflowSnapshot = nativeBinding.JsWorkflowSnapshot
 module.exports.XaiProvider = nativeBinding.XaiProvider
 module.exports.JsXaiProvider = nativeBinding.JsXaiProvider
+module.exports.addUsageToTokenUsage = nativeBinding.addUsageToTokenUsage
 module.exports.ChatRole = nativeBinding.ChatRole
 module.exports.JsChatRole = nativeBinding.JsChatRole
 module.exports.completeBatch = nativeBinding.completeBatch
 module.exports.completeBatchConfig = nativeBinding.completeBatchConfig
+module.exports.computeAudioCost = nativeBinding.computeAudioCost
 module.exports.computeElidSimilarity = nativeBinding.computeElidSimilarity
 module.exports.computeEmbeddingSimhashSimilarity = nativeBinding.computeEmbeddingSimhashSimilarity
+module.exports.computeImageCost = nativeBinding.computeImageCost
 module.exports.computeTextSimhashSimilarity = nativeBinding.computeTextSimhashSimilarity
+module.exports.computeVideoCost = nativeBinding.computeVideoCost
 module.exports.countMessageTokens = nativeBinding.countMessageTokens
+module.exports.defaultHttpClientConfig = nativeBinding.defaultHttpClientConfig
 module.exports.envVarForProvider = nativeBinding.envVarForProvider
 module.exports.estimateTokens = nativeBinding.estimateTokens
 module.exports.extractInlineArtifacts = nativeBinding.extractInlineArtifacts
+module.exports.FINISH_WORKFLOW_TOOL_NAME = nativeBinding.FINISH_WORKFLOW_TOOL_NAME
+module.exports.finishWorkflowTool = nativeBinding.finishWorkflowTool
+module.exports.finishWorkflowToolDef = nativeBinding.finishWorkflowToolDef
 module.exports.formatProviderHttpTail = nativeBinding.formatProviderHttpTail
 module.exports.getContextWindow = nativeBinding.getContextWindow
 module.exports.HistoryEventKindTag = nativeBinding.HistoryEventKindTag
@@ -870,10 +892,16 @@ module.exports.loadClientTls = nativeBinding.loadClientTls
 module.exports.loadServerTls = nativeBinding.loadServerTls
 module.exports.lookupPricing = nativeBinding.lookupPricing
 module.exports.lookupStepBuilder = nativeBinding.lookupStepBuilder
+module.exports.Modality = nativeBinding.Modality
+module.exports.JsModality = nativeBinding.JsModality
+module.exports.newRetryStack = nativeBinding.newRetryStack
+module.exports.newUsageEvent = nativeBinding.newUsageEvent
 module.exports.PauseReason = nativeBinding.PauseReason
 module.exports.JsPauseReason = nativeBinding.JsPauseReason
 module.exports.peerEnvelopeVersion = nativeBinding.peerEnvelopeVersion
 module.exports.peerTokenEnv = nativeBinding.peerTokenEnv
+module.exports.ProgressKind = nativeBinding.ProgressKind
+module.exports.JsProgressKind = nativeBinding.JsProgressKind
 module.exports.providerEnvVars = nativeBinding.providerEnvVars
 module.exports.ProviderId = nativeBinding.ProviderId
 module.exports.JsProviderId = nativeBinding.JsProviderId
@@ -886,6 +914,7 @@ module.exports.registerPricing = nativeBinding.registerPricing
 module.exports.registerStepBuilder = nativeBinding.registerStepBuilder
 module.exports.resolveApiKey = nativeBinding.resolveApiKey
 module.exports.resolveBeerToken = nativeBinding.resolveBeerToken
+module.exports.resolveRetryStack = nativeBinding.resolveRetryStack
 module.exports.runAgent = nativeBinding.runAgent
 module.exports.runAgentWithCallback = nativeBinding.runAgentWithCallback
 module.exports.SessionPausePolicy = nativeBinding.SessionPausePolicy
@@ -900,6 +929,7 @@ module.exports.TemplateRole = nativeBinding.TemplateRole
 module.exports.JsTemplateRole = nativeBinding.JsTemplateRole
 module.exports.tryDeserializeEvent = nativeBinding.tryDeserializeEvent
 module.exports.typedToolSimple = nativeBinding.typedToolSimple
+module.exports.unlimitedHttpClientConfig = nativeBinding.unlimitedHttpClientConfig
 module.exports.version = nativeBinding.version
 
 // --- post-build: typed-error wrapping ---

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "blazen",
-  "version": "0.1.153",
+  "version": "0.1.154",
   "description": "Blazen - Event-driven AI workflow framework for Node.js/TypeScript",
   "main": "index.js",
   "types": "index.d.ts",
@@ -40,23 +40,36 @@
   ],
   "devDependencies": {
     "@napi-rs/cli": "^3.0.0",
-    "@biomejs/biome": "^2"
+    "@biomejs/biome": "^2",
+    "ava": "^6.4.1"
   },
   "engines": {
     "node": ">= 18"
   },
+  "ava": {
+    "files": [
+      "../../tests/node/test_*.mjs"
+    ],
+    "extensions": [
+      "mjs"
+    ],
+    "timeout": "120s",
+    "concurrency": 4,
+    "verbose": true
+  },
   "optionalDependencies": {
-    "@blazen-dev/blazen-linux-x64-gnu": "0.1.153",
-    "@blazen-dev/blazen-linux-x64-musl": "0.1.153",
-    "@blazen-dev/blazen-linux-arm64-gnu": "0.1.153",
-    "@blazen-dev/blazen-linux-arm64-musl": "0.1.153",
-    "@blazen-dev/blazen-darwin-arm64": "0.1.153",
-    "@blazen-dev/blazen-win32-x64-msvc": "0.1.153"
+    "@blazen-dev/blazen-linux-x64-gnu": "0.1.154",
+    "@blazen-dev/blazen-linux-x64-musl": "0.1.154",
+    "@blazen-dev/blazen-linux-arm64-gnu": "0.1.154",
+    "@blazen-dev/blazen-linux-arm64-musl": "0.1.154",
+    "@blazen-dev/blazen-darwin-arm64": "0.1.154",
+    "@blazen-dev/blazen-win32-x64-msvc": "0.1.154"
   },
   "scripts": {
     "build": "napi build --release --platform --features local-all,langfuse --js index.js && node scripts/post-build.mjs",
     "build:debug": "napi build --platform --features local-all,langfuse --js index.js && node scripts/post-build.mjs",
-    "test": "node --test ../../tests/node/test_workflow.mjs",
-    "test:smoke": "node --test ../../tests/node/test_llm_smoke.mjs"
+    "test": "ava",
+    "test:workflow": "ava ../../tests/node/test_workflow.mjs",
+    "test:smoke": "ava ../../tests/node/test_llm_smoke.mjs"
   }
 }