Since yesterday we are having issues with tools that are used by our AI Agents: tools aren’t completing due to a timeout of 40 seconds (executionTimeMs: 40062).
The tools are relative simple scenarios: a deep research request with Perplexity, which normally takes around 60-90 seconds. But this gets canceled now with a “status: fail” and “result: Tool execution status unknown”.
I tried to solve it by doing a custom ASYNC HTTP request to the API combined with a polling structure, but the execution limit seems to be at “tool level”. So when the tool takes up more than 40 seconds, it fails.
How can we increase this or solve this? And did this limit changed recently? Because we didn’t have the issue since we’ve been using this AI Agent on a daily base.
It sounds like your AI agent tools are running into a strict 40 second execution time limit per tool, which is causing failures on longer-running tasks like deep research. Even using async HTTP requests and polling doesn’t help, since the timeout is enforced at the tool level, not just at the request level. You may need to restructure your workflow consider splitting complex tasks into smaller, sequential tools or shifting heavy-lift operations to external processing endpoints that report back once complete.
This limit of 40 seconds per tool is new to me. We’ve been using the tool for a week and the Perplexity module often requires a runtime of +/- 60 seconds, which is not weird for a deep research request. It has been working fine until yesterday.
Plus: I don’t see any way of splitting one single request to Perplexity… I can do a async request which will inmediately returns the ID, but I MUST poll it with another module inside the AI Agent, which is returning and processing the results. And even this polling seems to be limited to 40 seconds.
AND: if a HTTP module has a timeout limit of 300 seconds, how is it possible to use this inside a tool when the tool itself is limited to 40 seconds? Very odd…
I have a similar issue. I’m trying to use the deep research model from OpenAI. And I can’t get a result because of the timeout. Any suggestion to approach this issue?
