Built a quick deep research clone with the new OpenAI Agents SDK! Surprisingly good for something I spent less than an hour on. Here it is in action (code+architecture in next tweet)
00:00
Rohan Mehta
64 posts
Rohan Mehta
@_rohanmehta
- First launch at OpenAI! Go try all of theseWe're launching new tools to help developers build reliable and powerful AI agents. 🤖🔧 Timestamps: 01:54 Web search 02:41 File search 03:22 Computer use 04:07 Responses API 10:17 Agents SDK
00:00 - MCP support now available in the openai agents SDK! Especially love the tracing dashboard integration
- Replying to @stuffyokodrawsYou can configure the environment from chatgpt.com/codex/settings…, and add a setup script that has internet access!
- Replying to @_rohanmehtaCode and sample outputs are here: github.com/openai/openai-… It uses a few openAI models + the web search tool to: 1. Plan the research 2. Search the web for a bunch of information, and summarize it in the context of the topic. 3. Write a research report.
- Replying to @swyxLet me know what feedback you have! Especially rough edges
- Replying to @_rohanmehtaWould love to add more features - RAG, PDF+image upload, code execution etc. PRs welcome!
- Replying to @Francyrad and @OpenAIDevsThere's nothing quite like the ChatGPT Deep Research. But the results of this demo are pretty good!
- Replying to @_rohanmehtaDocs here: openai.github.io/openai-agents-…
- Replying to @raveeshbhallaAny suggestions for what docs/examples would be most useful?
- Replying to @hwchase17IMO the intent here isn't "graphs are bad", it's needing to learn a DSL to declare your graph that isn't great (e.g. using custom conditionals/loops etc instead of just regular programming)
- Small quality of life upgrade: "Hey Siri, ask ChatGPT ...". It actually works.
- Replying to @ozy2ez @OpenAIDevs and @OpenAIYes! You can define a function tool that calls any API you like. Docs: openai.github.io/openai-agents-…
