Transcripts hold the words of everyone who was recorded, your clients included. You are responsible for telling people they are being recorded, and for getting any consent the law where you work requires, before you record. You are also responsible for having the right to share those recordings and transcripts with the tools you connect.
Asking for today's walks
Ask your AI tool something like "what walks did I do today?". It calls list_walks for today and gets back your walks and daily sweeps: when each was recorded, the job or jobs it covers, the walk type and job phase, how long it is, how many takes it has, and whether it has a transcript, audio and video. A transcript still being made, or one that failed, says so. A daily sweep that covered several jobs is listed under each job it touched.
The list carries no transcript text and no money. On the public API it is GET /v1/walks.
Reading a transcript in pages
An hour's walk is long, so get_transcript returns a transcript in pages of time-stamped segments. Your tool asks for the next page with the cursor it was given. It can also read one stretch of one take: takeId picks the take, and fromSeconds and toSeconds pick the stretch, counted from the start of that take. A translated walk reads in the language Forge shows by default, or in the language it was recorded in with language.
On the public API it is GET /v1/walks/{id}/transcript.
Audio and video links
The first page of a transcript also lists each audio and video take, with a link to it. The link is for you, or for your own software, to open or download. Your AI tool works from the transcript.
- Each link works for 15 minutes.
- Forge records a read when it gives out a link, not when or whether the recording is played.
- Ending a connection stops everything at its very next request, except a link already given out, which keeps working until its 15 minutes are up. So audio and video links already handed out stop working within 15 minutes of a revoke.
On the public API the links come from GET /v1/walks/{id}/recordings.
What a sweep's transcript holds
A daily sweep's transcript comes whole. Each segment is labelled with the job it was matched to, or as unmatched, and carries its category, the decision a person made about it, and the to-do it proposed.
Numbers said out loud
Numbers spoken on a walk, costs included, appear in the transcript as they were said. The homeowner view of an estimate leaves costs out, but a transcript has no homeowner view, so check anything your tool drafts for a homeowner from a transcript.
Recipe: the kitchen photos for a job
Photos taken on a walk are filed to the job's areas, such as the kitchen, and to trades, from what was said at the time.
- Ask for the job's photos in one area:
get_jobwithphotoAreaset to the area's name or label, such as kitchen, in any case. The answer lists the job's areas and a page of photos from that area. An area that matches nothing gives an empty page and the list of areas, so your tool can ask again. - Look at each photo with
get_photoand the photo's id. - Photos a person hid are left out. To include them, ask again with
includeHiddenset to true, and read a hidden photo withincludeHiddenset to true.
On the public API the same filter is the area parameter of GET /v1/jobs/{id}/photos.
A photo's area and trades are filed automatically from what was said on the walk, unless a person filed it. Each photo says whether a person has confirmed its filing, and gives a confidence band. The band is not a measured accuracy, and this manual gives no figure for how often automatic filing is right.