Bring Your Own Model
Fifteen provider adapters and a separate binding for each thing the platform asks a model to do, so transcription and post-call reading are two decisions rather than one. The bindings are set by the operator, not inside a workspace.
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.
A purpose is a category and a modality together
Thirteen categories across four modalities, each bound on its own, so transcription and post-call reading never have to sit on the same model. Two of the thirteen are a cheap text reading and a heavier one, which is how short work and long work end up on different models and different bills.
One primary, then an ordered chain
Each purpose names a primary and a list of fallbacks in the order to try them. An unusable primary leaves the purpose unbound rather than promoting a fallback into the gap, because promoting one would hide the breakage.
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 jittered wait, then onward. Everything else advances immediately with no second try, and an authentication or missing deployment failure raises an alert on its way past rather than moving quietly.
Three strikes shuts a binding for half a minute
Failures are counted per provider and model in shared state, so an outage is noticed once instead of once per workspace. After the wait exactly one request is let through, and it either reopens the binding or restarts the clock. If that state is unreachable, 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 catalogue 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 afterwards.
More in Intelligence
12 capabilitiesAI that proposes edits to the record - an insight becomes a field once you accept it.
The model proposes record updates with its confidence and the quote it heard them in, shown as a before-and-after you accept or reject row by row.
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.
Positive, neutral or negative on every analysed call, stored in a column of its own so the ones that went wrong are a filter rather than a listening exercise.
Chosen words rather than a fixed list for how the agent sounded, how the customer did, and the conversation overall - the manner behind the sentiment.
Communication, resolution, satisfaction and compliance scored nought to ten beside a separately judged overall, on every call that clears the analysis gates.
Compliance scored nought to ten against what the vertical says matters, because no phrase list exists to recite. A missing reading is excluded, never passed.
What the customer was actually asking for and what the agent committed to, extracted as structured intents rather than left inside the transcript.
Callbacks and meetings pulled out with their times, a confidence figure and the words they came from, ready to accept in one click or reject in one.
Threads summarised and read for intent and urgency, but only when a participant matches a record, so the model is never called on mail that is not yours.
Draft from an instruction or rework a message you have, landing in the composer with a suggested subject and going nowhere at all until you send it.
Describe what you track or paste a spreadsheet's columns, and get a named field group back with types, options and placeholders to approve one at a time.
A sentence becomes a grid filter and a sort order with the reading explained back, on the toolbar of every grid that carries a column menu of its own.
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