Your first pipeline¶
We will sort a few tickets, keep the most urgent one, and return just the fields we need. Then we will run the same work as a flow. No model or provider is involved.
Before starting, install Ribbit. These commands use Bash on Linux/macOS or Git Bash on Windows. Save files as UTF-8. Start in an empty working directory.
Create three records¶
cat > tickets.jsonl <<'JSONL'
{"ticket":"R1","priority":3,"body":"Checkout fails."}
{"ticket":"R2","priority":1,"body":"A spelling mistake."}
{"ticket":"R3","priority":2,"body":"Keyboard checkout is blocked."}
JSONL
Read the first two lines as JSON values:
Expected output:
{"ticket":"R1","priority":3,"body":"Checkout fails."}
{"ticket":"R2","priority":1,"body":"A spelling mistake."}
--input jsonl tells Ribbit that each line is one record. --output jsonl gives us ordinary JSON values to inspect.
Build a pipe¶
ribbit sort --by priority --type number --descending \
--file tickets.jsonl --input jsonl |
ribbit take 1 |
ribbit select ticket,body --output jsonl
Expected output:
The sort uses the numeric priority we supplied. It does not infer urgency from the text. The intermediate commands keep Ribbit records; only the last command exports bare values.
Change take 1 to take 2 and run the pipeline again. R3 should appear after R1. You have changed the amount of output without changing the input file.
Run the same work as a flow¶
ribbit flow run --file tickets.jsonl --input jsonl --output jsonl -- \
sort --by priority --type number --descending :: \
take 1 :: select ticket,body
You should get the same R1 output. The :: tokens separate command steps inside a single Ribbit process.
Run it once more with --stats before --. The JSON result still goes to stdout; the statistics go to stderr and report zero inference requests.
You now have a pipeline you can change one step at a time. Continue with a local model, or read how records and pipes compose.