@datalayer/agent-runtimes 1.0.4 → 1.0.6

package/lib/types/evals.d.ts

@@ -1,5 +1,5 @@
 /**
- * Evaluation benchmark specification.
+ * Built-in evaluator specification.
  */
 export interface EvalSpec {
     /** Unique eval identifier */
@@ -8,20 +8,26 @@ export interface EvalSpec {
     version?: string;
     /** Display name */
     name: string;
-    /** Description of the evaluation */
+    /** Description of the evaluator */
     description: string;
-    /** Category: Coding, Knowledge, Reasoning, Agentic, or Safety */
-    category: 'Coding' | 'Knowledge' | 'Reasoning' | 'Agentic' | 'Safety';
-    /** Number of tasks in the benchmark */
-    task_count: number;
-    /** Primary metric (e.g., 'pass@1', 'accuracy', 'success_rate') */
-    metric: string;
-    /** Source URL or repository */
+    /** Evaluator family */
+    category: 'Comparison' | 'Type Validation' | 'Performance' | 'LLM-as-a-Judge' | 'Span-Based' | 'Report';
+    /** Case-level or report-level evaluator */
+    evaluator_type: 'case' | 'report';
+    /** Pydantic evaluator class name */
+    pydantic_class: string;
+    /** Primary output shape */
+    output_kind: 'boolean' | 'boolean_with_reason' | 'score' | 'score_and_assertion' | 'report_table' | 'report_curve';
+    /** Cost tier for running this evaluator */
+    cost_tier: 'free' | 'llm';
+    /** Expected latency profile */
+    latency: 'instant' | 'fast' | 'slow';
+    /** Runtime requirements (e.g. expected_output, logfire, model) */
+    requires?: string[];
+    /** Source documentation URL */
     source: string;
-    /** Difficulty level */
-    difficulty: 'easy' | 'medium' | 'hard' | 'expert';
-    /** Relevant languages */
-    languages: string[];
+    /** Suggested baseline configuration */
+    default_config?: Record<string, unknown>;
 }
 /**
  * Eval configuration for an agent spec.
@@ -31,11 +37,14 @@ export interface AgentEvalConfig {
     name?: string;
     description?: string;
     category?: string;
-    task_count?: number;
-    metric?: string;
+    evaluator_type?: string;
+    pydantic_class?: string;
+    output_kind?: string;
+    cost_tier?: string;
+    latency?: string;
+    requires?: string[];
     source?: string;
-    difficulty?: string;
-    languages?: string[];
+    default_config?: Record<string, unknown>;
     [key: string]: unknown;
 }
 export interface EvalReport {

package/lib/types/index.d.ts

@@ -4,6 +4,7 @@ export * from './ag-ui';
 export * from './agents';
 export * from './agents-lifecycle';
 export * from './agentspecs';
+export * from './benchmarks';
 export * from './chat';
 export * from './checkpoints';
 export * from './config';
@@ -26,9 +27,11 @@ export * from './middleware';
 export * from './models';
 export * from './notifications';
 export * from './outputs';
+export * from './personas';
 export * from './protocol';
 export * from './sandbox';
 export * from './skills';
+export * from './stream';
 export * from './teams';
 export * from './tool-approvals';
 export * from './tools';

package/lib/types/index.js

@@ -8,6 +8,7 @@ export * from './ag-ui';
 export * from './agents';
 export * from './agents-lifecycle';
 export * from './agentspecs';
+export * from './benchmarks';
 export * from './chat';
 export * from './checkpoints';
 export * from './config';
@@ -30,9 +31,11 @@ export * from './middleware';
 export * from './models';
 export * from './notifications';
 export * from './outputs';
+export * from './personas';
 export * from './protocol';
 export * from './sandbox';
 export * from './skills';
+export * from './stream';
 export * from './teams';
 export * from './tool-approvals';
 export * from './tools';

package/lib/types/mcp.d.ts

@@ -73,6 +73,11 @@ export interface McpServerStatus {
     status: 'none' | 'not_started' | 'starting' | 'failed' | 'started';
     error?: string;
     tools_count?: number;
+    tools?: Array<{
+        name: string;
+        description?: string;
+        enabled?: boolean;
+    }>;
 }
 /**
  * Full response from `/api/v1/configure/mcp-toolsets-status`.
@@ -84,6 +89,9 @@ export interface McpToolsetsStatusResponse {
     ready_servers: string[];
     failed_servers: Record<string, string>;
     servers: McpServerStatus[];
+    enabled_tools_by_server?: Record<string, string[]>;
+    enabled_tools_count?: number;
+    approved_tools_by_server?: Record<string, string[]>;
 }
 /**
  * Primer-addon indicator colours for each aggregate MCP status.

package/lib/types/models.d.ts

@@ -3,7 +3,7 @@ import type { MCPServer } from './mcp';
  * Specification for an AI model from the catalog.
  */
 export interface AIModel {
-    /** Unique model identifier (e.g., 'anthropic:claude-sonnet-4-5-20250514') */
+    /** Unique model identifier (e.g., 'anthropic:claude-3-5-haiku-20241022') */
     id: string;
     /** Model spec version */
     version: string;
@@ -22,7 +22,7 @@ export interface AIModel {
  * Configuration for an AI model runtime (as returned by the server).
  */
 export interface AIModelRuntime {
-    /** Model identifier (e.g., 'anthropic:claude-sonnet-4-5') */
+    /** Model identifier (e.g., 'anthropic:claude-3-5-haiku-20241022') */
     id: string;
     /** Display name for the model */
     name: string;

package/lib/types/personas.d.ts

@@ -0,0 +1,25 @@
+/**
+ * Specification for a Persona.
+ *
+ * A Persona is a lightweight identity built on top of an agent spec —
+ * it bundles a name, a description and a set of tags that describe the
+ * role and tone of the underlying agent.
+ */
+export interface PersonaSpec {
+    /** Unique persona identifier (e.g., 'tutor', 'sentinel') */
+    id: string;
+    /** Persona spec version */
+    version: string;
+    /** Display name of the persona */
+    name: string;
+    /** Short persona description */
+    description: string;
+    /** Categorization tags */
+    tags: string[];
+    /** Icon identifier */
+    icon?: string;
+    /** Emoji representation */
+    emoji?: string;
+    /** Optional reference to the underlying agent spec id */
+    agent?: string;
+}

package/lib/types/personas.js

@@ -0,0 +1,5 @@
+/*
+ * Copyright (c) 2025-2026 Datalayer, Inc.
+ * Distributed under the terms of the Modified BSD License.
+ */
+export {};

package/lib/types/skills.d.ts

@@ -1,5 +1,13 @@
 /**
- * Skill information from backend
+ * Skill status as tracked by the server-side skills area.
+ *
+ * - `available`: in catalog, not yet enabled
+ * - `enabled`: enabled, SKILL.md loading pending
+ * - `loaded`: SKILL.md loaded, included in the LLM system prompt
+ */
+export type SkillStatus = 'available' | 'enabled' | 'loaded';
+/**
+ * Skill information from the WebSocket snapshot.
  */
 export interface SkillInfo {
     id: string;
@@ -9,6 +17,40 @@ export interface SkillInfo {
     tags?: string[];
     has_scripts?: boolean;
     has_resources?: boolean;
+    /** Lifecycle status from the server-side skills area */
+    status?: SkillStatus;
+    /** Whether the skill is approved (defaults to true when omitted) */
+    approved?: boolean;
+    /** Raw SKILL.md content (populated once status is 'loaded') */
+    skill_definition?: string | null;
+    /** Discovery/source variant for this skill */
+    source_variant?: 'module' | 'package' | 'path' | 'unknown';
+    module?: string;
+    package?: string;
+    method?: string;
+    path?: string;
+}
+/**
+ * Loaded skill normalized for UI/state consumption.
+ *
+ * This representation is resolved from an agent's active spec and optional
+ * catalog enrichment, then persisted per-agent in the runtime store.
+ */
+export interface LoadedSkillInfo {
+    id: string;
+    name: string;
+    description: string;
+    variant: 'module' | 'package' | 'path' | 'unknown';
+    module?: string;
+    package?: string;
+    method?: string;
+    path?: string;
+    license?: string;
+    compatibility?: string;
+    allowedTools?: string[];
+    skillMetadata?: Record<string, string>;
+    tags?: string[];
+    emoji?: string;
 }
 /**
  * Skills response from backend

package/lib/types/stream.d.ts

@@ -0,0 +1,110 @@
+import type { ContextSnapshotData } from './context';
+import type { McpToolsetsStatusResponse } from './mcp';
+import type { SkillStatus } from './skills';
+export type AgentStreamEventType = 'agent.snapshot' | 'tool_approval_created' | 'tool_approval_approved' | 'tool_approval_rejected';
+export interface AgentStreamMessage<TPayload = Record<string, unknown>> {
+    version: string;
+    type: AgentStreamEventType | string;
+    agentId?: string;
+    timestamp: string;
+    payload: TPayload;
+    event?: string;
+    data?: unknown;
+}
+export interface AgentStreamToolApprovalPayload {
+    id: string;
+    agent_id?: string;
+    pod_name?: string;
+    tool_name: string;
+    tool_call_id?: string;
+    tool_args?: Record<string, unknown>;
+    status?: string;
+    note?: string | null;
+    created_at?: string;
+    updated_at?: string;
+}
+export interface AgentStreamSnapshotPayload {
+    agentId?: string;
+    approvals: AgentStreamToolApprovalPayload[];
+    pendingApprovalCount: number;
+    contextSnapshot?: ContextSnapshotData | null;
+    costUsage?: ContextSnapshotData['costUsage'];
+    mcpStatus?: McpToolsetsStatusResponse | null;
+    codemodeStatus?: CodemodeStatusData | null;
+    fullContext?: Record<string, unknown> | null;
+    graphTelemetry?: GraphTelemetryData | null;
+}
+/** Graph-level telemetry data from pydantic-graph execution. */
+export interface GraphTelemetryData {
+    agentId?: string;
+    graphName?: string | null;
+    /** Static topology: graph node definitions. */
+    nodes: GraphTelemetryNode[];
+    /** Static topology: edges between nodes. */
+    edges: GraphTelemetryEdge[];
+    /** Dynamic execution trace: per-node events. */
+    events: GraphNodeEvent[];
+    totalNodesExecuted: number;
+    totalDurationMs: number;
+    lastRunStartMs: number;
+    lastRunEndMs: number;
+    runCount: number;
+}
+export interface GraphTelemetryNode {
+    id: string;
+    name: string;
+    category: string;
+}
+export interface GraphTelemetryEdge {
+    source: string;
+    target: string;
+    label?: string | null;
+    edgeType: string;
+}
+export interface GraphNodeEvent {
+    nodeId: string;
+    nodeType: string;
+    status: string;
+    timestampMs: number;
+    durationMs?: number | null;
+    parentNodeId?: string | null;
+    error?: string | null;
+}
+/** Codemode status as pushed via the monitoring WebSocket. */
+export interface CodemodeStatusData {
+    enabled: boolean;
+    skills: Array<{
+        id?: string;
+        name: string;
+        description?: string;
+        tags?: string[];
+        has_scripts?: boolean;
+        has_resources?: boolean;
+        status?: SkillStatus;
+        approved?: boolean;
+        skill_definition?: string | null;
+        source_variant?: 'module' | 'package' | 'path' | 'unknown';
+        module?: string;
+        package?: string;
+        method?: string;
+        path?: string;
+    }>;
+    available_skills: Array<{
+        id?: string;
+        name: string;
+        description?: string;
+        tags?: string[];
+        has_scripts?: boolean;
+        has_resources?: boolean;
+        status?: SkillStatus;
+        approved?: boolean;
+        skill_definition?: string | null;
+        source_variant?: 'module' | 'package' | 'path' | 'unknown';
+        module?: string;
+        package?: string;
+        method?: string;
+        path?: string;
+    }>;
+    sandbox?: Record<string, unknown> | null;
+}
+export declare function parseAgentStreamMessage(raw: unknown): AgentStreamMessage | null;

package/lib/types/stream.js

@@ -0,0 +1,36 @@
+/*
+ * Copyright (c) 2025-2026 Datalayer, Inc.
+ * Distributed under the terms of the Modified BSD License.
+ */
+const isObject = (value) => !!value && typeof value === 'object';
+export function parseAgentStreamMessage(raw) {
+    if (!isObject(raw)) {
+        return null;
+    }
+    // New envelope format.
+    if (typeof raw.type === 'string' && isObject(raw.payload)) {
+        return {
+            version: typeof raw.version === 'string' ? raw.version : '1.0',
+            type: raw.type,
+            agentId: typeof raw.agentId === 'string' ? raw.agentId : undefined,
+            timestamp: typeof raw.timestamp === 'string'
+                ? raw.timestamp
+                : new Date().toISOString(),
+            payload: raw.payload,
+            event: typeof raw.event === 'string' ? raw.event : undefined,
+            data: raw.data,
+        };
+    }
+    // Legacy event/data format.
+    if (typeof raw.event === 'string' && isObject(raw.data)) {
+        return {
+            version: '1.0',
+            type: raw.event,
+            timestamp: new Date().toISOString(),
+            payload: raw.data,
+            event: raw.event,
+            data: raw.data,
+        };
+    }
+    return null;
+}

package/lib/utils/utils.d.ts

@@ -47,11 +47,15 @@ export declare function convertHistoryToDisplayItems(messages: ChatMessage[]): D
  */
 export declare function formatTokenCount(tokens: number): string;
 /**
- * Derive the API base URL from a configEndpoint.
- * configEndpoint may look like:
- *   "http://host:port/api/v1/config"     (from agentRuntimeConfig)
- *   "http://host:port/api/v1/configure"   (from Chat component)
- * We strip the trailing path segment to get "http://host:port/api/v1".
+ * Derives the API base URL from a config endpoint URL.
+ *
+ * The config endpoint may be `{base}/api/v1/config` (from agentRuntimeConfig)
+ * or `{base}/api/v1/configure` (from the Chat component). This function
+ * strips the trailing path segment to yield `{base}/api/v1`.
+ *
+ * @param configEndpoint - Full config endpoint URL.
+ *
+ * @returns The API base URL without the trailing config segment.
  */
 export declare function getApiBaseFromConfig(configEndpoint: string): string;
 /**

package/lib/utils/utils.js

@@ -166,11 +166,15 @@ export function formatTokenCount(tokens) {
     return tokens.toString();
 }
 /**
- * Derive the API base URL from a configEndpoint.
- * configEndpoint may look like:
- *   "http://host:port/api/v1/config"     (from agentRuntimeConfig)
- *   "http://host:port/api/v1/configure"   (from Chat component)
- * We strip the trailing path segment to get "http://host:port/api/v1".
+ * Derives the API base URL from a config endpoint URL.
+ *
+ * The config endpoint may be `{base}/api/v1/config` (from agentRuntimeConfig)
+ * or `{base}/api/v1/configure` (from the Chat component). This function
+ * strips the trailing path segment to yield `{base}/api/v1`.
+ *
+ * @param configEndpoint - Full config endpoint URL.
+ *
+ * @returns The API base URL without the trailing config segment.
  */
 export function getApiBaseFromConfig(configEndpoint) {
     return configEndpoint.replace(/\/(config|configure)\/?$/, '');

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@datalayer/agent-runtimes",
-  "version": "1.0.4",
+  "version": "1.0.6",
   "type": "module",
   "workspaces": [
     ".",
@@ -60,18 +60,19 @@
     "build:lib": "npm run clean:lib && gulp resources-to-lib && tsc -b",
     "build:nextjs": "npm run build --workspace=nextjs-notebook-example",
     "build:types": "npm run clean:lib && tsc -b",
-    "check": "npm run format:check && npm run lint:all && npm run type-check:all",
+    "check": "npm run format:check && npm run lint:all && npm run type-check:all && npm run typedoc",
     "check:fix": "npm run format:all && npm run lint:fix && npm run type-check:all",
     "clean": "rimraf lib dist build tsconfig.tsbuildinfo",
     "clean:cache": "rimraf node_modules/.vite",
     "clean:dist": "rimraf dist",
     "clean:lib": "rimraf lib tsconfig.tsbuildinfo",
     "create:patches": "bash scripts/create-patches.sh",
-    "examples": "run-p server:start jupyter:start examples:vite",
+    "examples": "run-p server:start server:start:codemode jupyter:start examples:vite",
+    "examples:proxy": "make examples:proxy",
     "examples:fresh": "npm run clean:cache && npm run examples",
-    "examples:jupyter": "run-p jupyter:start server:start examples:vite",
+    "examples:jupyter": "run-p jupyter:start server:start server:start:codemode examples:vite",
     "examples:nextjs": "npm run dev --workspace=nextjs-notebook-example",
-    "examples:vite": "VITE_APP_TARGET=examples VITE_DATALAYER_RUN_URL=http://localhost:8888 vite",
+    "examples:vite": "VITE_DATALAYER_RUN_URL=${VITE_DATALAYER_RUN_URL:-https://prod1.datalayer.run} VITE_DATALAYER_RUNTIMES_URL=${VITE_DATALAYER_RUNTIMES_URL:-https://r1.datalayer.run} VITE_DATALAYER_AGENT_RUNTIMES_URL=${VITE_DATALAYER_AGENT_RUNTIMES_URL:-https://r1.datalayer.run} VITE_APP_TARGET=examples vite",
     "format": "prettier --write \"src/**/*.{js,jsx,ts,tsx,css,json,md}\" \"examples/**/*.{js,jsx,ts,tsx,css,json,md,mjs}\"",
     "format:all": "npm run format && npm run format --workspaces --if-present",
     "format:check": "prettier --check \"src/**/*.{js,jsx,ts,tsx,css,json,md}\" \"examples/**/*.{js,jsx,ts,tsx,css,json,md,mjs}\"",
@@ -89,13 +90,18 @@
     "preview": "vite preview",
     "rebuild:fresh": "npm run create:patches && npm install && npm run build && npm run clean:cache",
     "server": "python -m agent_runtimes serve",
-    "server:start": "PYTHONIOENCODING=utf-8 AGENT_RUNTIMES_DEBUG=false AGENT_RUNTIMES_LOG_LEVEL=info python -m agent_runtimes serve --port 8765 --log-level info",
+    "server:start": "PYTHONIOENCODING=utf-8 AGENT_RUNTIMES_DEBUG=false AGENT_RUNTIMES_LOG_LEVEL=info DATALAYER_EVALS_MODE=interactive DATALAYER_EVALS_EMIT_LIVE_EVENTS=true python -m agent_runtimes serve --port 8765 --log-level info",
+    "server:start:agent-node": "PYTHONIOENCODING=utf-8 AGENT_RUNTIMES_DEBUG=false AGENT_RUNTIMES_LOG_LEVEL=info DATALAYER_EVALS_MODE=interactive DATALAYER_EVALS_EMIT_LIVE_EVENTS=true AGENT_RUNTIMES_INFERENCE_PROVIDER_OVERRIDE=${AGENT_RUNTIMES_INFERENCE_PROVIDER_OVERRIDE:-datalayer} python -m agent_runtimes serve --node --port 8765 --log-level info",
+    "server:start:codemode": "PYTHONIOENCODING=utf-8 AGENT_RUNTIMES_DEBUG=false AGENT_RUNTIMES_LOG_LEVEL=info DATALAYER_EVALS_MODE=interactive DATALAYER_EVALS_EMIT_LIVE_EVENTS=true python -m agent_runtimes serve --port 8766 --log-level info",
+    "examples:codemode": "run-p server:start server:start:codemode examples:vite",
     "start": "vite",
     "start:acp": "run-p server:start start:acp:vite",
     "start:acp:prod": "run-p server start:acp:vite",
     "start:acp:vite": "VITE_ACP_WS_URL=ws://localhost:8765/api/v1/acp/ws vite",
     "start:agent": "run-p server:start start:agent:vite",
     "start:agent:vite": "vite --open /html/agent.html",
+    "start:agent-node": "run-p server:start:agent-node start:agent-node:vite",
+    "start:agent-node:vite": "vite --open /html/agent-node.html",
     "start:agent-notebook": "run-p server:start jupyter:start start:agent-notebook:vite",
     "start:agent-notebook:vite": "vite --open /html/agent-notebook.html",
     "start:agent-lexical": "run-p server:start jupyter:start start:agent-lexical:vite",
@@ -120,7 +126,7 @@
   },
   "dependencies": {
     "@a2a-js/sdk": "^0.3.6",
-    "@a2ui/react": "^0.8.0",
+    "@a2ui/react": "^0.9.1",
     "@ag-ui/client": "^0.0.42",
     "@ag-ui/core": "^0.0.42",
     "@ag-ui/encoder": "^0.0.42",
@@ -128,12 +134,12 @@
     "@agentclientprotocol/sdk": "^0.8.0",
     "@ai-sdk/react": "^4.0.0-beta.33",
     "@anthropic-ai/sdk": "^0.52.0",
-    "@datalayer/core": "^1.0.13",
-    "@datalayer/icons-react": "^1.0.6",
+    "@datalayer/core": "^1.0.22",
+    "@datalayer/icons-react": "^1.0.7",
     "@datalayer/jupyter-lexical": "^1.0.16",
     "@datalayer/jupyter-react": "^2.0.7",
-    "@datalayer/primer-addons": "^1.0.10",
-    "@datalayer/primer-rjsf": "^1.0.1",
+    "@datalayer/primer-addons": "^1.0.15",
+    "@datalayer/primer-rjsf": "^1.0.3",
     "@excalidraw/excalidraw": "^0.18.0",
     "@jupyter-widgets/base-manager": "^1.0.12",
     "@jupyter-widgets/schema": "^0.5.6",
@@ -301,6 +307,7 @@
     "vitest": "^3.2.4"
   },
   "resolutions": {
+    "@langchain/core": "^0.3.80",
     "@ai-sdk/react": "^4.0.0-beta.33",
     "@microsoft/fast-colors": "5.3.1",
     "@microsoft/fast-components": "2.30.6",
@@ -318,6 +325,7 @@
     "typescript": "^5.8.3"
   },
   "overrides": {
+    "@langchain/core": "^0.3.80",
     "@ai-sdk/react": "^4.0.0-beta.33",
     "@microsoft/fast-colors": "5.3.1",
     "@microsoft/fast-components": "2.30.6",