Feature Area

Agent capabilities

Is your feature request related to a an existing bug? Please link it here.

While developing custom tooling, or adjusting settings in a small area within a large Crew it would be very useful to be able to rerun the entire job without invoking actual network calls. Running CrewAI offline would allow to evaluate changes that don't impact LLM output.

1. Faster iteration
2. Runs offline without networking
3. Predictable results / can be shared / used in tests
4. Does not use up tokens and lowers cost during development

Describe the solution you'd like

This could be enabled by executing the job once with crewai run --record flag and later with crewai run --replay

Describe alternatives you've considered

1. Could be done on Agent level where the agent notices the run mode and answers a cached value instead of executing its task
2. Could be done on LLM specification where Agent acts correctly, but LLM caches results
3. Could be done on networking layer which would store HAR files

Additional context

What exactly is saved on disk needs to be strongly considered. HAR files would need to be sanitized for auth information etc.

Willingness to Contribute

Yes, I'd be happy to submit a pull request