Sequa
Deterministic recording & replay for AI applications
5 followers
Deterministic recording & replay for AI applications
5 followers
Speed up integration tests for LLM applications from seconds to milliseconds. Record once, replay forever with Sequa, the open-source LLM cassette library.
Love the concept of recorded cassettes for LLM tests — would save so much flaky test pain. One thing I'd love to see is deterministic seeding or a "mock mode" that injects the recorded response but still lets you assert on token counts or latency budgets. That way you're not just replaying blind but can catch performance regressions too.
@nevzato5qw Thanks! Really appreciate the thoughtful feedback.
You're spot on, recording and replaying is just one part of the testing workflow. We've been thinking about making the replay layer more configurable, where you can replay recorded responses while still validating execution metadata like latency, token usage, and cost budgets.
That would make Sequa useful not only for deterministic functional tests but also for regression and performance testing. Definitely something we'd like to explore as the project evolves.
Thanks for the suggestion!