Replay without a recorder¶

Record mode is for the first run. Once you have the recordings, you want the opposite: no waiting, and an error for anything unknown, so a test never silently invents an answer.

Switch a project to replay¶

Mode is a setting of each project, and semula keeps it across restarts:

docker exec semula semula mode replay --project default

default: replay (kept across restarts)

Now ask something that was never recorded:

curl -i http://127.0.0.1:8080/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Name a vegetable."}]}'

HTTP/1.1 424 Failed Dependency

{"nearest":{"entry_id":"01M5HGZ4Y2JJHS5HCTKW5E12GT","score":0.667,"last_message":"Name a fruit."},"error":{"message":"No recorded answer matches this request, and semula is in replay mode.","type":"semula_error","param":null,"code":"semula_no_match"}}

The call fails at once with status 424. The body names the closest recorded prompt and its score, which is usually enough to see why a prompt you expected to match did not.

Recorded prompts keep working, with no network and no agent:

curl -s http://127.0.0.1:8080/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model": "gpt-4o", "messages": [{"role": "user", "content": "Name a fruit."}]}'

Two levels of mode¶

semula mode with no arguments shows both:

docker exec semula semula mode record --project default
docker exec semula semula mode

default: record (kept across restarts)
server default   record
project default  record

The server default starts from SEMULA_MODE and is forgotten on restart. A project that has not chosen its own mode follows it. A project's own choice sticks. Use --project for anything you want to survive a restart.

Next: let an agent do the answering.