Skip to content

0.6.0

Choose a tag to compare

@github-actions github-actions released this 15 Nov 20:53
· 16 commits to main since this release
d5a62d6

Added

  • stats calls show counts the stored calls. It gives the calls by direction and outcome, per employee, by hour
    and by day, the missed incoming calls and whether the client was called back, and the short calls. It reads only
    the local store. See a report of the week.
  • search meetings finds a call by what was said. By words, and with --model embeddinggemma by meaning too,
    with a text model on this computer. meetings embed --yes builds the search data ahead of time, one batch per run.
    models list and models download now also offer the text models. See
    find calls by what was meant.
  • tags add, tags remove and tags list mark calls. tags add complaint --matching "refund OR complaint"
    marks every call the search finds, and a rerun marks only the new ones; --meeting <id> marks one call. See
    mark every call about refunds.
  • A database that listens only on the phone system is reached through its SSH. Set db_ssh_host, db_ssh_user
    and a key file in db_ssh_key_path, or a password; the server's key is pinned in db_ssh_host_key first. pbx
    opens the tunnel for each command and closes it after. See
    a database behind the phone system's SSH.
  • A profile can keep more than one password. A database profile can sign in to SFTP or SSH with a password too.
    session start asks each password it needs; piped, one per line. They are kept in the profile's one keyring
    entry.
  • Cloud transcription, for one profile at a time. config set transcribe_engine openai sends the speech of the
    recordings to OpenAI in place of the local model, with the key in OPENAI_API_KEY. The run says where the audio
    goes before the first piece leaves. It stays off unless the profile's own settings turn it on. See
    cloud transcription.
  • meetings show prints a WireCat Desktop link for the call in the terminal, wirecat://open/v1/meeting/…. It
    opens the call once Desktop supports links to calls.
  • meetings show names each recording's file on this computer in local_path, so WireCat Desktop can play it.
    It is null when the recording is read over SFTP. This comes with cli-core 0.668.0.
  • Master.csv and the recordings of any size over SFTP. They are read in pieces, and a connection that breaks is
    opened again where it stopped. A recording on the phone system is copied into a private temporary file for the
    time it is read, and deleted after. A WAV, GSM or M4A recording of any size is transcribed, up to an hour of audio.
  • A stereo WAV, GSM or M4A recording keeps its two channels, so recordings_channels names the caller and the
    employee on each line for these formats too.

Changed — may break callers

  • A stereo WAV, GSM or M4A call is transcribed as two channels. Its lines say Channel 1 and Channel 2, or the
    names recordings_channels gives, where they were one mixed text. meetings transcribe --replace does it for the
    calls already transcribed.
  • An MP3, Ogg or FLAC recording can be up to 64 MB, not 256 MB. That is more than an hour at usual phone quality,
    and a recording longer than an hour was never transcribed.
  • The run record names the database's requests sql.connect, sql.columns and sql.rows, not cdr.connect,
    cdr.columns and cdr.rows. A database that does not answer in time now has the reason db.timeout, and one that
    cannot be reached network.failed.