feat: WIP add evals and telemetry abstractions

packages/core/package.json

@@ -32,6 +32,7 @@
     "test:unit": "vitest run"
   },
   "dependencies": {
+    "@opentelemetry/api": "^1.9.0",
     "@sindresorhus/is": "^7.0.0",
     "dedent": "^1.5.3",
     "delay": "^6.0.0",

packages/core/src/telemetry/index.ts

@@ -0,0 +1,2 @@
+export * from './noop-tracer'
+export * from './telemetry'

packages/core/src/telemetry/noop-tracer.ts

@@ -0,0 +1,81 @@
+import type { Span, SpanContext, Tracer } from '@opentelemetry/api'
+
+/**
+ * Tracer implementation that does nothing.
+ */
+export const noopTracer: Tracer = {
+  startSpan(): Span {
+    return noopSpan
+  },
+
+  startActiveSpan<F extends (span: Span) => unknown>(
+    _name: unknown,
+    arg1: unknown,
+    arg2?: unknown,
+    arg3?: F
+  ): any {
+    if (typeof arg1 === 'function') {
+      return arg1(noopSpan)
+    }
+
+    if (typeof arg2 === 'function') {
+      return arg2(noopSpan)
+    }
+
+    if (typeof arg3 === 'function') {
+      return arg3(noopSpan)
+    }
+  }
+}
+
+const noopSpan: Span = {
+  spanContext() {
+    return noopSpanContext
+  },
+
+  setAttribute() {
+    return this
+  },
+
+  setAttributes() {
+    return this
+  },
+
+  addEvent() {
+    return this
+  },
+
+  addLink() {
+    return this
+  },
+
+  addLinks() {
+    return this
+  },
+
+  setStatus() {
+    return this
+  },
+
+  updateName() {
+    return this
+  },
+
+  end() {
+    return this
+  },
+
+  isRecording() {
+    return false
+  },
+
+  recordException() {
+    return this
+  }
+}
+
+const noopSpanContext: SpanContext = {
+  traceId: '',
+  spanId: '',
+  traceFlags: 0
+}

packages/core/src/telemetry/telemetry.ts

@@ -0,0 +1,188 @@
+import {
+  type Attributes,
+  type AttributeValue,
+  type Span,
+  type SpanOptions,
+  SpanStatusCode,
+  trace,
+  type Tracer
+} from '@opentelemetry/api'
+
+import type * as types from '../types'
+import { noopTracer } from './noop-tracer'
+
+export type AgenticSpanOptions = {
+  attributes?: {
+    [attributeKey: string]:
+      | AttributeValue
+      | { input: () => AttributeValue | undefined }
+      | { output: () => AttributeValue | undefined }
+      | undefined
+  }
+}
+
+export class Telemetry {
+  public readonly isEnabled: boolean
+  public readonly tracer: Tracer
+  public readonly recordInputs: boolean
+  public readonly recordOutputs: boolean
+  public readonly metadata: Record<string, AttributeValue>
+
+  constructor({
+    tracer,
+    isEnabled = true,
+    recordInputs = true,
+    recordOutputs = true,
+    metadata = {}
+  }: {
+    tracer?: Tracer
+
+    /**
+     * Enable or disable telemetry. Disabled by default.
+     */
+    isEnabled?: boolean
+
+    /**
+     * Enable or disable input recording. Enabled by default.
+     *
+     * You might want to disable input recording to avoid recording sensitive
+     * information, to reduce data transfers, or to increase performance.
+     */
+    recordInputs?: boolean
+
+    /**
+     * Enable or disable output recording. Enabled by default.
+     *
+     * You might want to disable output recording to avoid recording sensitive
+     * information, to reduce data transfers, or to increase performance.
+     */
+    recordOutputs?: boolean
+
+    /**
+     * Additional information to include in the telemetry data.
+     */
+    metadata?: Record<string, AttributeValue>
+  }) {
+    this.isEnabled = !!isEnabled
+    this.tracer =
+      tracer ?? (this.isEnabled ? trace.getTracer('agentic') : noopTracer)
+    this.recordInputs = recordInputs
+    this.recordOutputs = recordOutputs
+    this.metadata = metadata
+  }
+
+  recordSpan<T>(
+    {
+      name,
+      attributes = {},
+      endWhenDone = true,
+      ...spanOptions
+    }: {
+      name: string
+      endWhenDone?: boolean
+    } & Omit<SpanOptions, 'attributes'> &
+      AgenticSpanOptions,
+    implementation: (span: Span) => types.MaybePromise<T>
+  ): Promise<T> {
+    const spanAttributes = this.convertAttributes({ attributes })
+
+    return this.tracer.startActiveSpan(
+      name,
+      {
+        ...spanOptions,
+        attributes: spanAttributes
+      },
+      async (span) => {
+        try {
+          const result: Awaited<T> = await Promise.resolve(implementation(span))
+
+          if (endWhenDone) {
+            span.end()
+          }
+
+          return result
+        } catch (err) {
+          try {
+            if (err instanceof Error) {
+              span.recordException({
+                name: err.name,
+                message: err.message,
+                stack: err.stack
+              })
+
+              span.setStatus({
+                code: SpanStatusCode.ERROR,
+                message: err.message
+              })
+            } else {
+              span.setStatus({ code: SpanStatusCode.ERROR })
+            }
+          } finally {
+            // Always end the span when there is an error.
+            span.end()
+          }
+
+          throw err
+        }
+      }
+    )
+  }
+
+  convertAttributes({ attributes = {} }: AgenticSpanOptions): Attributes {
+    return {
+      ...Object.fromEntries(
+        Object.entries(attributes)
+          .map(([key, value]) => {
+            if (value === undefined) {
+              return [key, value]
+            }
+
+            // input value, check if it should be recorded:
+            if (
+              typeof value === 'object' &&
+              'input' in value &&
+              typeof value.input === 'function'
+            ) {
+              if (!this.recordInputs) {
+                return undefined
+              }
+
+              const result = value.input()
+              if (result === undefined) {
+                return undefined
+              } else {
+                return [key, result]
+              }
+            }
+
+            // output value, check if it should be recorded:
+            if (
+              typeof value === 'object' &&
+              'output' in value &&
+              typeof value.output === 'function'
+            ) {
+              if (!this.recordOutputs) {
+                return undefined
+              }
+
+              const result = value.output()
+              if (result === undefined) {
+                return undefined
+              } else {
+                return [key, result]
+              }
+            }
+
+            return [key, value]
+          })
+          .filter(Boolean)
+      ),
+
+      ...Object.fromEntries(
+        Object.entries(this.metadata).map(([key, value]) => {
+          return [`agentic.telemetry.metadata.${key}`, value]
+        })
+      )
+    }
+  }
+}

packages/evals/package.json

@@ -0,0 +1,48 @@
+{
+  "name": "@agentic/evals",
+  "version": "0.1.0",
+  "description": "TODO",
+  "author": "Travis Fischer <travis@transitivebullsh.it>",
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/transitive-bullshit/agentic.git"
+  },
+  "type": "module",
+  "source": "./src/index.ts",
+  "types": "./dist/index.d.ts",
+  "sideEffects": false,
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js",
+      "default": "./dist/index.js"
+    }
+  },
+  "files": [
+    "dist"
+  ],
+  "scripts": {
+    "build": "tsup --config ../../tsup.config.ts",
+    "dev": "tsup --config ../../tsup.config.ts --watch",
+    "clean": "del dist",
+    "test": "run-s test:*",
+    "test:lint": "eslint .",
+    "test:typecheck": "tsc --noEmit",
+    "test:unit": "vitest run"
+  },
+  "dependencies": {
+    "type-fest": "^4.21.0"
+  },
+  "peerDependencies": {
+    "@agentic/core": "workspace:*",
+    "zod": "^3.23.8"
+  },
+  "devDependencies": {
+    "@agentic/core": "workspace:*",
+    "@agentic/tsconfig": "workspace:*"
+  },
+  "publishConfig": {
+    "access": "public"
+  }
+}

packages/evals/src/index.ts

@@ -0,0 +1 @@
+// TODO