Bring Your Own Model
Fifteen providers, and every job picks its own, so transcription and post-call reading are two decisions rather than one. When one starts timing out mid-afternoon, the work moves down the chain rather than stopping. We wire that up, not you.
Which model answers this call
healthy- primaryAnthropic2/3 failuresserving
- fallback 1Azure OpenAIstandby
- fallback 2Bedrockstandby
Product figures from the platform’s own defaults - not customer averages
How it works.
Short work and long work do not have to share a model
Thirteen jobs across four kinds of work, each pointed at its own model, so transcription and post-call reading never have to sit on the same one. Two of the thirteen are a cheap reading and a heavier one, which is how short work and long work end up on different models and different bills.
One first choice, then an ordered line behind it
Each job names its first choice and the ones to try after it, in order. A first choice that cannot be used leaves that job unbound rather than quietly promoting the next one, because a quiet promotion is how you end up not knowing what read your calls.
What went wrong decides what happens next
A rate limit or a timeout counts as capacity: one further attempt on the same model behind a staggered wait, then onward. Everything else moves on at once with no second try, and a bad key or a missing deployment raises an alert on its way past rather than going by quietly.
Three strikes and it stops asking for thirty seconds
Failures are counted per provider and model in one shared place, so an outage is noticed once rather than once per workspace. After the wait exactly one request is let through, and it either reopens the door or restarts the clock. If that shared place cannot be reached, everything is treated as healthy.
One moment in every read.
Every read passes through the same seven. Bring Your Own Model is the lit one, and everything either side of it is a different page in this category.
- 01Source
the call or the thread it reads
- 02Transcribe
audio into words, with speakers
- 03Read
the pass over the whole of it
- 04Judge
the score, the sentiment, the intent
- 05Extract
the fields and follow-ups pulled out
- 06Write
what lands back on the record
- 07Review
a person checking the machine
The specifics.
8 facts- Coverage
- Fifteen adapters, nine text capable, over thirteen categories and four modalities
- Nothing bound
- Fails loudly and names what is missing. Audio routes alone may fall back to environment keys
- Chain
- A primary plus an ordered fallback list, per purpose. Not one global preference
- When one fails
- A rate limit or a timeout gets two attempts on that model, then the next. Everything else moves on at once. Configuration failures raise an alert as they pass, and the last one in the chain is thrown
- Breaker
- Three consecutive failures, thirty seconds shut, one probe to reopen it
- Timeout
- Sixty seconds per attempt, unless the calling code asks for its own limit
- Where it is bound
- One platform-level catalog of providers, models and purposes. Credentials are held once and encrypted; there is no per-workspace key store, so this is the operator's choice on your behalf
- Not the same as
- This is which model answers. Which instruction it gets is the presets page
What it reads, and what reads it.
Nothing here invents its input. These are where the material comes from, and where the verdict goes afterward.
More in Intelligence
12 capabilitiesAI that proposes edits to the record - an insight becomes a field once you accept it.
Nothing on an existing lead moves by itself. What the call gave up arrives as proposed edits, each with the value it would replace and the words it came from.
One to six sentences on the record covering why the call happened and what was agreed, so nobody reading the list has to open the audio to find out.
Finding yesterday's bad calls is a filter, not an afternoon. Positive, neutral or negative on every analyzed call, kept in a field of its own you can sort on.
How the agent sounded, how the customer sounded, one word for the pair. Chosen words rather than a fixed list, so a call can close politely and grudgingly.
A month of calls gets reviewed by whoever had time. Every call carries five numbers instead, and the overall is judged in its own right rather than averaged.
Nobody re-listens to check the disclosure went out. Every call is scored against what compliance means on your floor, with a missing one excluded, not passed.
What the customer asked for and what your agent promised, kept as two lists, because they are two different obligations and only one of them is yours to keep.
A promise made out loud is worth nothing until it sits on a day. Callbacks and meetings come off the call with the time said, real the moment you accept.
A thread is read only when somebody on it is one of your leads. No match and nothing is sent anywhere, so mail that is not about a customer is left alone.
A draft or a rewrite that lands in the composer and stops there. The subject is a suggestion, every word editable, and nothing leaves until a person sends it.
The field group you would otherwise build one field at a time. Describe what you track or paste a spreadsheet's columns, then approve or refuse each field.
Type the list you want and it builds the filter and the sort. Then it says in one sentence what it understood: a filter nobody typed has to say what it did.
The rest of the platform.
Five more categories, all on the same record and the same bill. Each card names three of its capabilities, so you can tell from here whether it is worth opening.
Intelligence
See bring your own model on your own floor.
Thirty minutes, your numbers and your data. We will set bring your own model up live and you can decide from the thing itself rather than from this page.
14-day trial · no card · migration included