@axlsdk/studio 0.14.0 → 0.15.0

package/dist/server/index.d.cts

@@ -1,35 +1,164 @@
 import * as hono_ws from 'hono/ws';
+import { C as ConnectionManager, S as StudioEnv, a as CostData } from '../connection-manager-BMPahDuY.cjs';
+export { B as BroadcastTarget } from '../connection-manager-BMPahDuY.cjs';
 import * as hono_types from 'hono/types';
 import { Hono } from 'hono';
-import { AxlRuntime } from '@axlsdk/axl';
-import { C as ConnectionManager, a as CostData, S as StudioEnv } from '../connection-manager-B7AWpsCD.cjs';
-export { B as BroadcastTarget } from '../connection-manager-B7AWpsCD.cjs';
+import { AxlRuntime, TraceEvent, ExecutionInfo, EvalHistoryEntry } from '@axlsdk/axl';
 
 /**
- * Accumulates cost data from trace events.
- * Broadcasts updates to the 'costs' WS channel.
+ * Pure reducer functions for each aggregate panel.
+ * Each reducer is a pure (state, source) => state function — no I/O, no mutation.
  */
-declare class CostAggregator {
-    private connMgr;
-    private data;
-    constructor(connMgr: ConnectionManager);
-    /** Process a trace event and update cost data. */
-    onTrace(event: {
-        type?: string;
-        agent?: string;
-        model?: string;
-        workflow?: string;
-        cost?: number;
-        tokens?: {
-            input?: number;
-            output?: number;
-            reasoning?: number;
-        };
-    }): void;
-    /** Get current aggregated cost data. */
-    getData(): CostData;
-    /** Reset all accumulated data. */
-    reset(): void;
+
+type EvalTrendRun = {
+    timestamp: number;
+    id: string;
+    scores: Record<string, number>;
+    cost: number;
+    /** Primary model for this run (first entry of `metadata.models`). Undefined
+     *  when the run has no recorded models (e.g., legacy data or test harnesses). */
+    model?: string;
+    /** Total run duration in ms (from `EvalResult.duration`). */
+    duration?: number;
+};
+type EvalTrendEntry = {
+    runs: EvalTrendRun[];
+    latestScores: Record<string, number>;
+    scoreMean: Record<string, number>;
+    scoreStd: Record<string, number>;
+    costTotal: number;
+    runCount: number;
+};
+type EvalTrendData = {
+    byEval: Record<string, EvalTrendEntry>;
+    totalRuns: number;
+    totalCost: number;
+};
+type WorkflowStatsData = {
+    byWorkflow: Record<string, {
+        total: number;
+        completed: number;
+        failed: number;
+        /** Bounded sorted array of recent durations for p50/p95. Max MAX_DURATIONS entries. */
+        durations: number[];
+        durationSum: number;
+        avgDuration: number;
+    }>;
+    totalExecutions: number;
+    failureRate: number;
+};
+type TraceStatsData = {
+    eventTypeCounts: Record<string, number>;
+    byTool: Record<string, {
+        calls: number;
+        denied: number;
+        approved: number;
+    }>;
+    retryByAgent: Record<string, {
+        schema: number;
+        validate: number;
+        guardrail: number;
+    }>;
+    totalEvents: number;
+};
+
+type WindowId = '24h' | '7d' | '30d' | 'all';
+type AggregateBroadcast<State> = {
+    snapshots: Record<WindowId, State>;
+    updatedAt: number;
+};
+
+type TraceReducer<State> = (acc: State, event: TraceEvent) => State;
+type TraceAggregatorOptions<State> = {
+    runtime: AxlRuntime;
+    connMgr: ConnectionManager;
+    channel: string;
+    reducer: TraceReducer<State>;
+    emptyState: () => State;
+    windows: WindowId[];
+    /** Max executions to replay on rebuild. Default 2000. */
+    executionCap?: number;
+    /** Optional transform applied to each window's state before WS broadcast. */
+    broadcastTransform?: (state: State) => unknown;
+};
+/**
+ * Consumes TraceEvents from execution history and the live trace stream.
+ * Maintains per-window aggregate snapshots via a pure reducer.
+ */
+declare class TraceAggregator<State> {
+    private snaps;
+    private interval?;
+    private listener?;
+    private options;
+    constructor(options: TraceAggregatorOptions<State>);
+    start(): Promise<void>;
+    rebuild(): Promise<void>;
+    getSnapshot(window: WindowId): State;
+    getAllSnapshots(): Record<WindowId, State>;
+    close(): void;
+}
+
+type ExecutionReducer<State> = (acc: State, execution: ExecutionInfo) => State;
+type ExecutionAggregatorOptions<State> = {
+    runtime: AxlRuntime;
+    connMgr: ConnectionManager;
+    channel: string;
+    reducer: ExecutionReducer<State>;
+    emptyState: () => State;
+    windows: WindowId[];
+    /** Max executions to replay on rebuild. Default 2000. */
+    executionCap?: number;
+    /** Optional transform applied to each window's state before WS broadcast. */
+    broadcastTransform?: (state: State) => unknown;
+};
+/**
+ * Consumes ExecutionInfo at the execution granularity (not individual trace events).
+ * Live updates arrive via workflow_end trace events — the aggregator fetches the
+ * finalized ExecutionInfo and folds it.
+ */
+declare class ExecutionAggregator<State> {
+    private snaps;
+    private interval?;
+    private listener?;
+    private options;
+    /** Generation counter to prevent stale async fold after rebuild. */
+    private generation;
+    constructor(options: ExecutionAggregatorOptions<State>);
+    start(): Promise<void>;
+    rebuild(): Promise<void>;
+    getSnapshot(window: WindowId): State;
+    getAllSnapshots(): Record<WindowId, State>;
+    close(): void;
+}
+
+type EvalReducer<State> = (acc: State, entry: EvalHistoryEntry) => State;
+type EvalAggregatorOptions<State> = {
+    runtime: AxlRuntime;
+    connMgr: ConnectionManager;
+    channel: string;
+    reducer: EvalReducer<State>;
+    emptyState: () => State;
+    windows: WindowId[];
+    /** Max eval entries to replay on rebuild. Default 500. */
+    entryCap?: number;
+    /** Optional transform applied to each window's state before WS broadcast. */
+    broadcastTransform?: (state: State) => unknown;
+};
+/**
+ * Consumes EvalHistoryEntry. Rebuilds from runtime.getEvalHistory().
+ * Live updates arrive via runtime.on('eval_result', entry).
+ */
+declare class EvalAggregator<State> {
+    private snaps;
+    private interval?;
+    private listener?;
+    private options;
+    constructor(options: EvalAggregatorOptions<State>);
+    start(): Promise<void>;
+    rebuild(): Promise<void>;
+    getSnapshot(window: WindowId): State;
+    getAllSnapshots(): Record<WindowId, State>;
+    close(): void;
 }
 
 type CreateServerOptions = {
@@ -48,7 +177,11 @@ type CreateServerOptions = {
 declare function createServer(options: CreateServerOptions): {
     app: Hono<StudioEnv, hono_types.BlankSchema, "/">;
     connMgr: ConnectionManager;
-    costAggregator: CostAggregator;
+    costAggregator: TraceAggregator<CostData>;
+    workflowStatsAggregator: ExecutionAggregator<WorkflowStatsData>;
+    traceStatsAggregator: TraceAggregator<TraceStatsData>;
+    evalTrendsAggregator: EvalAggregator<EvalTrendData>;
+    aggregatorStartPromise: Promise<void | [void, void, void, void]>;
     /** Create WS handlers. Call before registering static/SPA routes are reached. */
     createWsHandlers: () => {
         onOpen(_event: Event, ws: hono_ws.WSContext): void;
@@ -57,6 +190,10 @@ declare function createServer(options: CreateServerOptions): {
         onError(_event: Event, ws: hono_ws.WSContext): void;
     };
     traceListener: (event: unknown) => void;
+    /** Abort all active streaming eval runs. */
+    closeActiveRuns: () => void;
+    /** Close all aggregators (clear intervals and unsubscribe listeners). */
+    closeAggregators: () => void;
 };
 
-export { ConnectionManager, CostAggregator, type CreateServerOptions, StudioEnv, createServer };
+export { type AggregateBroadcast, ConnectionManager, type CreateServerOptions, EvalAggregator, ExecutionAggregator, StudioEnv, TraceAggregator, type WindowId, createServer };

package/dist/server/index.d.ts

@@ -1,35 +1,164 @@
 import * as hono_ws from 'hono/ws';
+import { C as ConnectionManager, S as StudioEnv, a as CostData } from '../connection-manager-BMPahDuY.js';
+export { B as BroadcastTarget } from '../connection-manager-BMPahDuY.js';
 import * as hono_types from 'hono/types';
 import { Hono } from 'hono';
-import { AxlRuntime } from '@axlsdk/axl';
-import { C as ConnectionManager, a as CostData, S as StudioEnv } from '../connection-manager-B7AWpsCD.js';
-export { B as BroadcastTarget } from '../connection-manager-B7AWpsCD.js';
+import { AxlRuntime, TraceEvent, ExecutionInfo, EvalHistoryEntry } from '@axlsdk/axl';
 
 /**
- * Accumulates cost data from trace events.
- * Broadcasts updates to the 'costs' WS channel.
+ * Pure reducer functions for each aggregate panel.
+ * Each reducer is a pure (state, source) => state function — no I/O, no mutation.
  */
-declare class CostAggregator {
-    private connMgr;
-    private data;
-    constructor(connMgr: ConnectionManager);
-    /** Process a trace event and update cost data. */
-    onTrace(event: {
-        type?: string;
-        agent?: string;
-        model?: string;
-        workflow?: string;
-        cost?: number;
-        tokens?: {
-            input?: number;
-            output?: number;
-            reasoning?: number;
-        };
-    }): void;
-    /** Get current aggregated cost data. */
-    getData(): CostData;
-    /** Reset all accumulated data. */
-    reset(): void;
+
+type EvalTrendRun = {
+    timestamp: number;
+    id: string;
+    scores: Record<string, number>;
+    cost: number;
+    /** Primary model for this run (first entry of `metadata.models`). Undefined
+     *  when the run has no recorded models (e.g., legacy data or test harnesses). */
+    model?: string;
+    /** Total run duration in ms (from `EvalResult.duration`). */
+    duration?: number;
+};
+type EvalTrendEntry = {
+    runs: EvalTrendRun[];
+    latestScores: Record<string, number>;
+    scoreMean: Record<string, number>;
+    scoreStd: Record<string, number>;
+    costTotal: number;
+    runCount: number;
+};
+type EvalTrendData = {
+    byEval: Record<string, EvalTrendEntry>;
+    totalRuns: number;
+    totalCost: number;
+};
+type WorkflowStatsData = {
+    byWorkflow: Record<string, {
+        total: number;
+        completed: number;
+        failed: number;
+        /** Bounded sorted array of recent durations for p50/p95. Max MAX_DURATIONS entries. */
+        durations: number[];
+        durationSum: number;
+        avgDuration: number;
+    }>;
+    totalExecutions: number;
+    failureRate: number;
+};
+type TraceStatsData = {
+    eventTypeCounts: Record<string, number>;
+    byTool: Record<string, {
+        calls: number;
+        denied: number;
+        approved: number;
+    }>;
+    retryByAgent: Record<string, {
+        schema: number;
+        validate: number;
+        guardrail: number;
+    }>;
+    totalEvents: number;
+};
+
+type WindowId = '24h' | '7d' | '30d' | 'all';
+type AggregateBroadcast<State> = {
+    snapshots: Record<WindowId, State>;
+    updatedAt: number;
+};
+
+type TraceReducer<State> = (acc: State, event: TraceEvent) => State;
+type TraceAggregatorOptions<State> = {
+    runtime: AxlRuntime;
+    connMgr: ConnectionManager;
+    channel: string;
+    reducer: TraceReducer<State>;
+    emptyState: () => State;
+    windows: WindowId[];
+    /** Max executions to replay on rebuild. Default 2000. */
+    executionCap?: number;
+    /** Optional transform applied to each window's state before WS broadcast. */
+    broadcastTransform?: (state: State) => unknown;
+};
+/**
+ * Consumes TraceEvents from execution history and the live trace stream.
+ * Maintains per-window aggregate snapshots via a pure reducer.
+ */
+declare class TraceAggregator<State> {
+    private snaps;
+    private interval?;
+    private listener?;
+    private options;
+    constructor(options: TraceAggregatorOptions<State>);
+    start(): Promise<void>;
+    rebuild(): Promise<void>;
+    getSnapshot(window: WindowId): State;
+    getAllSnapshots(): Record<WindowId, State>;
+    close(): void;
+}
+
+type ExecutionReducer<State> = (acc: State, execution: ExecutionInfo) => State;
+type ExecutionAggregatorOptions<State> = {
+    runtime: AxlRuntime;
+    connMgr: ConnectionManager;
+    channel: string;
+    reducer: ExecutionReducer<State>;
+    emptyState: () => State;
+    windows: WindowId[];
+    /** Max executions to replay on rebuild. Default 2000. */
+    executionCap?: number;
+    /** Optional transform applied to each window's state before WS broadcast. */
+    broadcastTransform?: (state: State) => unknown;
+};
+/**
+ * Consumes ExecutionInfo at the execution granularity (not individual trace events).
+ * Live updates arrive via workflow_end trace events — the aggregator fetches the
+ * finalized ExecutionInfo and folds it.
+ */
+declare class ExecutionAggregator<State> {
+    private snaps;
+    private interval?;
+    private listener?;
+    private options;
+    /** Generation counter to prevent stale async fold after rebuild. */
+    private generation;
+    constructor(options: ExecutionAggregatorOptions<State>);
+    start(): Promise<void>;
+    rebuild(): Promise<void>;
+    getSnapshot(window: WindowId): State;
+    getAllSnapshots(): Record<WindowId, State>;
+    close(): void;
+}
+
+type EvalReducer<State> = (acc: State, entry: EvalHistoryEntry) => State;
+type EvalAggregatorOptions<State> = {
+    runtime: AxlRuntime;
+    connMgr: ConnectionManager;
+    channel: string;
+    reducer: EvalReducer<State>;
+    emptyState: () => State;
+    windows: WindowId[];
+    /** Max eval entries to replay on rebuild. Default 500. */
+    entryCap?: number;
+    /** Optional transform applied to each window's state before WS broadcast. */
+    broadcastTransform?: (state: State) => unknown;
+};
+/**
+ * Consumes EvalHistoryEntry. Rebuilds from runtime.getEvalHistory().
+ * Live updates arrive via runtime.on('eval_result', entry).
+ */
+declare class EvalAggregator<State> {
+    private snaps;
+    private interval?;
+    private listener?;
+    private options;
+    constructor(options: EvalAggregatorOptions<State>);
+    start(): Promise<void>;
+    rebuild(): Promise<void>;
+    getSnapshot(window: WindowId): State;
+    getAllSnapshots(): Record<WindowId, State>;
+    close(): void;
 }
 
 type CreateServerOptions = {
@@ -48,7 +177,11 @@ type CreateServerOptions = {
 declare function createServer(options: CreateServerOptions): {
     app: Hono<StudioEnv, hono_types.BlankSchema, "/">;
     connMgr: ConnectionManager;
-    costAggregator: CostAggregator;
+    costAggregator: TraceAggregator<CostData>;
+    workflowStatsAggregator: ExecutionAggregator<WorkflowStatsData>;
+    traceStatsAggregator: TraceAggregator<TraceStatsData>;
+    evalTrendsAggregator: EvalAggregator<EvalTrendData>;
+    aggregatorStartPromise: Promise<void | [void, void, void, void]>;
     /** Create WS handlers. Call before registering static/SPA routes are reached. */
     createWsHandlers: () => {
         onOpen(_event: Event, ws: hono_ws.WSContext): void;
@@ -57,6 +190,10 @@ declare function createServer(options: CreateServerOptions): {
         onError(_event: Event, ws: hono_ws.WSContext): void;
     };
     traceListener: (event: unknown) => void;
+    /** Abort all active streaming eval runs. */
+    closeActiveRuns: () => void;
+    /** Close all aggregators (clear intervals and unsubscribe listeners). */
+    closeAggregators: () => void;
 };
 
-export { ConnectionManager, CostAggregator, type CreateServerOptions, StudioEnv, createServer };
+export { type AggregateBroadcast, ConnectionManager, type CreateServerOptions, EvalAggregator, ExecutionAggregator, StudioEnv, TraceAggregator, type WindowId, createServer };

package/dist/server/index.js

@@ -1,11 +1,15 @@
 import {
   ConnectionManager,
-  CostAggregator,
+  EvalAggregator,
+  ExecutionAggregator,
+  TraceAggregator,
   createServer
-} from "../chunk-HUKUQDYL.js";
+} from "../chunk-IPDMFFTQ.js";
 export {
   ConnectionManager,
-  CostAggregator,
+  EvalAggregator,
+  ExecutionAggregator,
+  TraceAggregator,
   createServer
 };
 //# sourceMappingURL=index.js.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@axlsdk/studio",
-  "version": "0.14.0",
+  "version": "0.15.0",
   "description": "Local development UI for debugging, testing, and iterating on Axl agents and workflows",
   "type": "module",
   "main": "./dist/server/index.cjs",
@@ -35,22 +35,26 @@
     "dist"
   ],
   "dependencies": {
-    "hono": "^4.7.0",
     "@hono/node-server": "^1.13.0",
     "@hono/node-ws": "^1.1.0",
+    "hono": "^4.7.0",
     "tsx": "^4.19.0",
     "ws": "^8.0.0",
-    "@axlsdk/axl": "0.14.0"
+    "@axlsdk/axl": "0.15.0"
   },
   "devDependencies": {
     "@tailwindcss/vite": "^4.1.0",
     "@tanstack/react-query": "^5.64.0",
+    "@testing-library/jest-dom": "^6.9.1",
+    "@testing-library/react": "^16.3.2",
+    "@testing-library/user-event": "^14.6.1",
     "@types/node": "^25.2.3",
-    "@types/ws": "^8.0.0",
     "@types/react": "^19.0.0",
     "@types/react-dom": "^19.0.0",
+    "@types/ws": "^8.0.0",
     "@vitejs/plugin-react": "^4.3.0",
     "concurrently": "^9.1.0",
+    "jsdom": "^29.0.2",
     "lucide-react": "^0.474.0",
     "react": "^19.0.0",
     "react-dom": "^19.0.0",
@@ -61,10 +65,10 @@
     "vite": "^6.0.0",
     "vitest": "^3.0.0",
     "zod": "^4.0.0",
-    "@axlsdk/testing": "0.14.0"
+    "@axlsdk/testing": "0.15.0"
   },
   "peerDependencies": {
-    "@axlsdk/eval": "0.14.0"
+    "@axlsdk/eval": "0.15.0"
   },
   "peerDependenciesMeta": {
     "@axlsdk/eval": {