NEW: OpenAI announces new tools for building agents.
Here is everything you need to know:
OpenAI has already launched two big agent solutions like Deep Research and Operator.
The tools are now coming to the APIs for developers to build their own agents.
The tools are now coming to the APIs for developers to build their own agents.
The first built-in tool is called the web search tool.
This allows the models to access information from the internet for up-to-date and factual responses. It's the same tool that powers ChatGPT search.
Powered by a fine-tuned model under the hood...
This allows the models to access information from the internet for up-to-date and factual responses. It's the same tool that powers ChatGPT search.
Powered by a fine-tuned model under the hood...
It uses a GPT-4o/mini that it's good at looking a web-retrieved data, finding relevant information, and citations.
GPT-4o search preview hits 90% on the SimpleQA benchmark. That's SoTA!
GPT-4o search preview hits 90% on the SimpleQA benchmark. That's SoTA!
The second tool is called the file search tool.
This is useful for agentic RAG related use cases.
It now supports metadata filtering and direct search endpoint which enables direct search to your vector databases.
This is useful for agentic RAG related use cases.
It now supports metadata filtering and direct search endpoint which enables direct search to your vector databases.
The third tool is the computer use tool.
This is like the Operator available via the APIs.
It allows you to control the computer you operate. This comes with the computer use model that's used by Operator.
This is like the Operator available via the APIs.
It allows you to control the computer you operate. This comes with the computer use model that's used by Operator.
This computer use tool achieves SoTA on several benchmarks.
They also announced the Responses API.
Unlike the traditional chat completions API, this new API is flexible enough to support multiple turns and tools more natively.
Responses API is a superset of what Chat Completion API supports.
Unlike the traditional chat completions API, this new API is flexible enough to support multiple turns and tools more natively.
Responses API is a superset of what Chat Completion API supports.
You can pair this endpoint with tools like file search.
This is a great way to bring in information about users into your applications.
This is a great way to bring in information about users into your applications.
You can also pair tools together with the Responses API.
It can call multiple tools at once and give you a final response in one request.
Responses API will be extended more and also support everything that the Assistant API (planned to be removed in 2026) can do.
It can call multiple tools at once and give you a final response in one request.
Responses API will be extended more and also support everything that the Assistant API (planned to be removed in 2026) can do.
The computer use tool can also be used with the Responses API.
You can add instructions and customize the display.
You can add instructions and customize the display.
OpenAI has also decided to make Swarm, their agent orchestration framework, more production-ready.
It has been rebranded to the Agents SDK. You can use the same tools with this framework. It uses the Responses API under the hood. Other vendors are also supported.
It has been rebranded to the Agents SDK. You can use the same tools with this framework. It uses the Responses API under the hood. Other vendors are also supported.
The Agents SDK, which is open-source, supports building multi-agents out of the box.
The triage agent can hand off tasks with the relevant context to execute tasks.
Using the Python SDK: pip install openai-agents
The Javascript version coming soon!
The triage agent can hand off tasks with the relevant context to execute tasks.
Using the Python SDK: pip install openai-agents
The Javascript version coming soon!
The Agents SDK supports monitoring and tracing out of the box.
This can be used for debugging your agents.
A tracing UI is also available to track traces of your agentic workflows.
Other features include guardrails and lifecycle events.
This can be used for debugging your agents.
A tracing UI is also available to track traces of your agentic workflows.
Other features include guardrails and lifecycle events.
Helpful links:
Agents Blog: openai.com/index/new-tool…
Built-in tools: platform.openai.com/docs/guides/to…
Responses API: platform.openai.com/docs/api-refer…
Agents SDK: github.com/openai/openai-…
Agents Blog: openai.com/index/new-tool…
Built-in tools: platform.openai.com/docs/guides/to…
Responses API: platform.openai.com/docs/api-refer…
Agents SDK: github.com/openai/openai-…
I am hosting a FREE webinar if you are interested in learning how to build effective agentic workflows with OpenAI's Agents SDK.
Join me here: dair-ai.thinkific.com/products/live_…
Join me here: dair-ai.thinkific.com/products/live_…
Generated by Thread Navigator
Press ⌘ + S to quick-export