One project per app¶

Recordings are grouped into projects, so two apps with the same prompt can have different answers. The project default always exists.

docker exec semula semula project create support-bot --name "Support bot"
docker exec semula semula mode record --project support-bot

created project support-bot
support-bot: record (kept across restarts)

Pick a project per call¶

Choose the project with a header or with the URL. Both mean the same thing:

curl -H 'X-Semula-Project: support-bot' \
  http://127.0.0.1:8080/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model": "m", "messages": [{"role": "user", "content": "Hi"}]}'

curl http://127.0.0.1:8080/v1/projects/support-bot/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model": "m", "messages": [{"role": "user", "content": "Hi"}]}'

With the OpenAI SDK, set base_url="http://127.0.0.1:8080/v1/projects/support-bot" once and forget it.

Author an entry before any call¶

You do not have to wait for a miss. entry add writes a request and its answer up front:

docker exec semula semula entry add support-bot \
  --prompt "Classify: my screen is cracked" \
  --text '{"label": "hardware"}'

created entry 01M5HH1FQC65G543SK7JAABCJV (version 1)

For a request with a system prompt, tools or parameters, put the whole chat body in a file with --request-file and the answer in --response-file. --model and --system cover the common small cases.

The entry is there now, with no hits yet:

docker exec semula semula entry list support-bot

01M5HH1FQC65G543SK7JAABCJV  v1  -   hits=0  -   aliases=0   Classify: my screen is cracked
(1 of 1)

Next: edit, pin and template.