Is it:

• Cost?

• Latency?

• Non-determinism?

• Something else?

Your feedback will directly shape what we build next.

One design decision we made early:

We didn't want developers rewriting their applications just to use Sequa.

The goal was:

Record.

We're just getting started.

Some areas we're actively exploring:

More framework integrations

Better storage backends

One thing kept frustrating us while building AI applications:

Every test run made another LLM API call.

That meant:

• Slower CI pipelines

• Higher API costs

• Flaky tests because model outputs changed

A common question is:

"What if my prompts contain sensitive information?"

Sequa includes optional PII masking before recordings are stored.

You choose what gets recorded.

Replay isn't the same as mocking.

When Sequa records an execution, it captures the real interaction between your application and the LLM.

Future runs replay that recorded execution, so your tests are:

• Faster

• Deterministic

• Offline

• Free from additional API calls

Current support includes:

OpenAI

LangChain

LangGraph