Transcribe a recorded audio file (meeting / call / standup) with the local whisper tool, then extract concrete action items and decisions with per-item confidence. Preview-first: shows the extracted items and writes nothing; set write_mode to create to save the confident, titled items as Recued tasks. Content-isolated via Gateway-gated file refs. Source: a local audio path or a data.file ref from another step.

How it works 22 steps

Inspect the data fetches, transforms, gates, and output this recipe runs.

Process (22 steps)

■ transcribed ?

skip: config.source is_empty

■ file_ref default

Apply default

■ has_file compare

Check if file ref has data

■ extract ?

skip: step.has_file not_equal true

■ action_items_raw default

Apply default

■ decisions default

Apply default

■ extraction_confidence default

Apply default

■ overall_below_threshold compare

Check if extraction confidence is less than setting confidence threshold

■ action_rows_raw map

Extract from each item

■ action_rows_lc map

Extract priority from each item

■ action_rows_pri map

Extract priority from each item

■ action_rows_with_due map

Extract due_date_iso from each item

■ action_rows_ready filter

Filter by condition

■ action_rows slice

Take a subset of

■ skip_writes any

Check if any of the conditions are true

■ create_tasks ?

skip: step.skip_writes equal true

■ action_count count

Count items in action rows

■ task_create_result_count count

Count items in create tasks

■ task_write_attempts ternary

Apply ternary

■ decisions_text join

Join items with " "

■ action_table to_table

Format results as a data table

■ card to_summary

Format results as a summary card

Settings 6 configurable

Configurable at install. Defaults shown — change them anytime in Recued.

title setting =

source setting =

max items setting = 12

write mode setting = preview

container names setting = [object Object]

confidence threshold setting = 0.7