🏔

examples/hf0-demo.ts

@@ -0,0 +1,83 @@
+import { OpenAIClient } from '@agentic/openai-fetch'
+import 'dotenv/config'
+import { z } from 'zod'
+
+import {
+  Agentic,
+  HumanFeedbackMechanismTwilio,
+  SearchAndCrawlTool,
+  WeatherTool,
+  withHumanFeedback
+} from '@/index'
+
+async function main() {
+  const openai = new OpenAIClient({ apiKey: process.env.OPENAI_API_KEY! })
+  const agentic = new Agentic({ openai })
+
+  const topic = process.argv[2] || 'OpenAI'
+
+  const res0 = await withHumanFeedback(
+    agentic
+      .gpt4({
+        messages: [
+          {
+            role: 'system',
+            content: `You are a MckInsey analyst who is an expert at writing executive summaries.`
+          },
+          {
+            role: 'user',
+            content: `What are the 3 most important questions we would need to answer in order to have a thorough understanding of this topic: {{topic}}? Be concise but creative in your questions, and make sure to capture the true essence of the topic.`
+          }
+        ],
+        model: 'gpt-4',
+        temperature: 1.0
+      })
+      .input(
+        z.object({
+          topic: z.string()
+        })
+      )
+      .output(z.array(z.string()).describe('question')),
+    {
+      type: 'selectN',
+      mechanism: HumanFeedbackMechanismTwilio
+    }
+  ).callWithMetadata({ topic })
+
+  console.log()
+  console.log()
+  console.log(`Questions: ${res0.result}`)
+  console.log()
+  console.log()
+
+  const questions: string[] = (res0.metadata.feedback as any)!.selected
+
+  const res = await agentic
+    .gpt4({
+      messages: [
+        {
+          role: 'system',
+          content: `You are a MckInsey analyst who is an expert at writing executive summaries.`
+        },
+        {
+          role: 'user',
+          content: `Write a thorough executive summary on this topic: {{topic}}.
+          In order to do this, you will need to answer the following questions: \n{{#questions}}- {{.}}\n{{/questions}}`
+        }
+      ],
+      model: 'gpt-4-32k'
+    })
+    .tools([new SearchAndCrawlTool(), new WeatherTool()])
+    .input(
+      z.object({
+        topic: z.string(),
+        questions: z.array(z.string())
+      })
+    )
+    .call({ topic, questions })
+
+  console.log('\n\n\n')
+  console.log(res)
+}
+
+main()

src/errors.ts

@@ -1,7 +1,11 @@
+import { TimeoutError as KyTimeoutError } from 'ky'
+import { TimeoutError } from 'p-timeout'
 import type { Jsonifiable } from 'type-fest'
 import type { ZodError } from 'zod'
 import { ValidationError, fromZodError } from 'zod-validation-error'
 
+export { TimeoutError, KyTimeoutError }
+
 export type ErrorOptions = {
   /** HTTP status code for the error. */
   status?: number

src/human-feedback/feedback.ts

@@ -216,6 +216,11 @@ export abstract class HumanFeedbackMechanism<
       choices.push(HumanFeedbackUserActions.Abort)
     }
 
+    this._agentic.logger.info(
+      choices,
+      `>>> Human feedback ${this.constructor.name} ${msg}`
+    )
+
     const choice =
       choices.length === 1
         ? HumanFeedbackUserActions.Select
@@ -291,6 +296,10 @@ export abstract class HumanFeedbackMechanism<
       })
     }
 
+    this._agentic.logger.info(
+      feedback,
+      `<<< Human feedback ${this.constructor.name} ${msg}`
+    )
     return feedback as FeedbackTypeToMetadata<T>
   }
 }

src/task.ts

@@ -192,6 +192,12 @@ export abstract class BaseTask<
           } else if (err instanceof errors.HumanFeedbackDeclineError) {
             ctx.retryMessage = err.message
             return
+          } else if (
+            err instanceof errors.KyTimeoutError ||
+            err instanceof errors.TimeoutError ||
+            (err as any).name === 'TimeoutError'
+          ) {
+            // TODO
           } else if ((err.cause as any)?.code === 'UND_ERR_HEADERS_TIMEOUT') {
             // TODO: This is a pretty common OpenAI error, and I think it either has
             // to do with OpenAI's servers being flaky or the combination of Node.js