kopia lustrzana https://github.com/transitive-bullshit/chatgpt-api
126 wiersze
3.1 KiB
Markdown
126 wiersze
3.1 KiB
Markdown
---
|
|
title: OpenAI Responses
|
|
description: How to use Agentic tools with the OpenAI Responses API.
|
|
---
|
|
|
|
<Tip>
|
|
There's no need for an adapter with the OpenAI SDK since all agentic tools are
|
|
compatible with OpenAI by default.
|
|
</Tip>
|
|
|
|
## Install
|
|
|
|
<CodeGroup>
|
|
```bash npm
|
|
npm install openai @agentic/platform-tool-client
|
|
```
|
|
|
|
```bash pnpm
|
|
pnpm add openai @agentic/platform-tool-client
|
|
```
|
|
|
|
```bash bun
|
|
bun add openai @agentic/platform-tool-client
|
|
```
|
|
|
|
```bash yarn
|
|
yarn add openai @agentic/platform-tool-client
|
|
```
|
|
|
|
</CodeGroup>
|
|
|
|
## Usage
|
|
|
|
This example uses the [`@agentic/search`](https://agentic.so/marketplace/projects/@agentic/search) tool.
|
|
|
|
```ts
|
|
import 'dotenv/config'
|
|
|
|
import { assert } from '@agentic/core'
|
|
import { AgenticToolClient } from '@agentic/platform-tool-client'
|
|
import OpenAI from 'openai'
|
|
|
|
async function main() {
|
|
const searchTool = await AgenticToolClient.fromIdentifier('@agentic/search')
|
|
const openai = new OpenAI()
|
|
|
|
const messages: OpenAI.Responses.ResponseInput = [
|
|
{
|
|
role: 'system',
|
|
content: 'You are a helpful assistant. Be as concise as possible.'
|
|
},
|
|
{ role: 'user', content: 'What is the weather in San Francisco?' }
|
|
]
|
|
|
|
{
|
|
// First call to OpenAI to invoke the tool
|
|
const res = await openai.responses.create({
|
|
model: 'gpt-4o-mini',
|
|
temperature: 0,
|
|
tools: searchTool.functions.responsesToolSpecs,
|
|
tool_choice: 'required',
|
|
input: messages
|
|
})
|
|
|
|
const toolCall = res.output[0]
|
|
assert(toolCall?.type === 'function_call')
|
|
const toolResult = await searchTool.callTool(
|
|
toolCall.name,
|
|
toolCall.arguments
|
|
)
|
|
|
|
messages.push(toolCall)
|
|
messages.push({
|
|
type: 'function_call_output',
|
|
call_id: toolCall.call_id,
|
|
output: JSON.stringify(toolResult)
|
|
})
|
|
}
|
|
|
|
{
|
|
// Second call to OpenAI to generate a text response
|
|
const res = await openai.responses.create({
|
|
model: 'gpt-4o-mini',
|
|
temperature: 0,
|
|
tools: searchTool.functions.responsesToolSpecs,
|
|
input: messages
|
|
})
|
|
|
|
console.log(res.output_text)
|
|
}
|
|
}
|
|
|
|
await main()
|
|
```
|
|
|
|
## Running this example
|
|
|
|
You can view the full source for this example here: https://github.com/transitive-bullshit/agentic/tree/main/examples/ts-sdks/openai
|
|
|
|
<Info>
|
|
You'll need an [OpenAI API key](https://platform.openai.com/docs/quickstart)
|
|
to run this example. Store it in a local `.env` file as `OPENAI_API_KEY`.
|
|
</Info>
|
|
|
|
<Info>
|
|
The
|
|
[`@agentic/search`](https://agentic.so/marketplace/projects/@agentic/search)
|
|
tool comes with a generous free tier, but once that runs out, you'll need to
|
|
sign up for a paid plan and add an `AGENTIC_API_KEY` to your `.env` file.
|
|
</Info>
|
|
|
|
```sh
|
|
git clone git@github.com:transitive-bullshit/agentic.git
|
|
cd agentic
|
|
pnpm install
|
|
pnpm build
|
|
echo 'OPENAI_API_KEY=your-key' >> .env
|
|
npx tsx examples/ts-sdks/openai/bin/weather-responses.ts
|
|
```
|
|
|
|
## Additional resources
|
|
|
|
- [OpenAI Chat Completions API docs](https://platform.openai.com/docs/api-reference/chat/create)
|
|
- [OpenAI Responses vs Chat Completions](https://platform.openai.com/docs/guides/responses-vs-chat-completions)
|
|
- [Using OpenAI's Chat Completion API with Agentic](/marketplace/ts-sdks/openai-chat)
|