01 /↗
An inbox that knows
where things belong.
Identify intent, urgency and escalation requests. Give every incoming message a useful next step.
Intelligence, woven into your business.What if your software
What if your software
just knew?
What needs attention. Where it belongs. What happens next.
Give Siltstone the context. Ask 1000s of questions at once.
Get structured answers that set your business in motion.
ONE PIECE OF CONTEXT. A WHOLE LOT OF CLARITY.
THE CONTEXT YOU ALREADY HAVE
Charged twice. Still waiting.
Ready to understandsiltstoneTap the stone
THE ANSWERS THAT MOVE YOU FORWARD
Is there a billing issue?Yes
Does this need attention today?Yes
Is an escalation requested?Yes
Which team should handle it?billing
4 answers. One request. All together.
NOW YOUR APPLICATION CANPrioritise. Route to billing. Flag for a manager.
Your support app knows where to send it.Under the surface Explore the API
POST /askRequest
{
"document": "We were charged twice and nobody has replied. Please get a manager to call me today.",
"threshold": 0.9,
"questions": [
{
"id": "billing_issue",
"type": "bool",
"ask": "Is there a billing issue?"
},
{
"id": "time_sensitive",
"type": "bool",
"ask": "Does this need attention today?"
},
{
"id": "escalation_requested",
"type": "bool",
"ask": "Is an escalation requested?"
},
{
"id": "team",
"type": "choice",
"ask": "Which team should handle it?",
"options": [
"billing",
"technical",
"sales"
]
}
]
}Response excerptIllustrative values
{
"threshold": 0.9,
"count": 4,
"flagged_for_review": 0,
"answers": [
{
"id": "billing_issue",
"type": "bool",
"value": true,
"confidence": 0.98,
"margin": 0.96,
"review": false,
"probabilities": {
"yes": 0.98,
"no": 0.02
}
},
{
"id": "time_sensitive",
"type": "bool",
"value": true,
"confidence": 0.98,
"margin": 0.96,
"review": false,
"probabilities": {
"yes": 0.98,
"no": 0.02
}
},
{
"id": "escalation_requested",
"type": "bool",
"value": true,
"confidence": 0.98,
"margin": 0.96,
"review": false,
"probabilities": {
"yes": 0.98,
"no": 0.02
}
},
{
"id": "team",
"type": "choice",
"value": "billing",
"confidence": 0.98,
"margin": 0.97,
"review": false,
"probabilities": {
"billing": 0.98,
"technical": 0.01,
"sales": 0.01
}
}
]
}Request and response fields follow the API. This example is illustrative, not live inference. The response excerpt omits runtime metadata and timings.
THE CONTEXT YOU ALREADY HAVE
The right model for the job.
Ready to understandsiltstoneTap the stone
THE ANSWERS THAT MOVE YOU FORWARD
Is this a translation request?Yes
Does it need multi-step reasoning?No
Does it need external tools?No
Which model fits these rules?fast
4 answers. One request. All together.
NOW YOUR APPLICATION CANSend the request to your fast model.
Your router uses the model this task calls for.Under the surface Explore the API
POST /askRequest
{
"document": "Request: “Translate hello into French.” Routing rules: use the fast model for short translations. Use the reasoning model for multi-step analysis.",
"threshold": 0.9,
"questions": [
{
"id": "translation",
"type": "bool",
"ask": "Is this a translation request?"
},
{
"id": "reasoning_needed",
"type": "bool",
"ask": "Does it need multi-step reasoning?"
},
{
"id": "tools_needed",
"type": "bool",
"ask": "Does it need external tools?"
},
{
"id": "model",
"type": "choice",
"ask": "Which model fits these rules?",
"options": [
"fast",
"reasoning"
]
}
]
}Response excerptIllustrative values
{
"threshold": 0.9,
"count": 4,
"flagged_for_review": 0,
"answers": [
{
"id": "translation",
"type": "bool",
"value": true,
"confidence": 0.98,
"margin": 0.96,
"review": false,
"probabilities": {
"yes": 0.98,
"no": 0.02
}
},
{
"id": "reasoning_needed",
"type": "bool",
"value": false,
"confidence": 0.98,
"margin": 0.96,
"review": false,
"probabilities": {
"yes": 0.02,
"no": 0.98
}
},
{
"id": "tools_needed",
"type": "bool",
"value": false,
"confidence": 0.98,
"margin": 0.96,
"review": false,
"probabilities": {
"yes": 0.02,
"no": 0.98
}
},
{
"id": "model",
"type": "choice",
"value": "fast",
"confidence": 0.98,
"margin": 0.96,
"review": false,
"probabilities": {
"fast": 0.98,
"reasoning": 0.02
}
}
]
}Request and response fields follow the API. This example is illustrative, not live inference. The response excerpt omits runtime metadata and timings.
THE CONTEXT YOU ALREADY HAVE
Check before the next step.
Ready to understandsiltstoneTap the stone
THE ANSWERS THAT MOVE YOU FORWARD
Is the proposed action destructive?Yes
Is it within the requested scope?No
Does the policy permit writing?No
What should happen next?review
4 answers. One request. All together.
NOW YOUR APPLICATION CANHold the action. Request a human review.
Your application enforces the policy.Under the surface Explore the API
POST /askRequest
{
"document": "Task: find duplicate files and report them. Proposed action: delete all matches. Policy: this task permits reading files only.",
"threshold": 0.9,
"questions": [
{
"id": "destructive",
"type": "bool",
"ask": "Is the proposed action destructive?"
},
{
"id": "within_scope",
"type": "bool",
"ask": "Is it within the requested scope?"
},
{
"id": "write_permitted",
"type": "bool",
"ask": "Does the policy permit writing?"
},
{
"id": "route",
"type": "choice",
"ask": "What should happen next?",
"options": [
"proceed",
"review"
]
}
]
}Response excerptIllustrative values
{
"threshold": 0.9,
"count": 4,
"flagged_for_review": 0,
"answers": [
{
"id": "destructive",
"type": "bool",
"value": true,
"confidence": 0.98,
"margin": 0.96,
"review": false,
"probabilities": {
"yes": 0.98,
"no": 0.02
}
},
{
"id": "within_scope",
"type": "bool",
"value": false,
"confidence": 0.98,
"margin": 0.96,
"review": false,
"probabilities": {
"yes": 0.02,
"no": 0.98
}
},
{
"id": "write_permitted",
"type": "bool",
"value": false,
"confidence": 0.98,
"margin": 0.96,
"review": false,
"probabilities": {
"yes": 0.02,
"no": 0.98
}
},
{
"id": "route",
"type": "choice",
"value": "review",
"confidence": 0.98,
"margin": 0.96,
"review": false,
"probabilities": {
"proceed": 0.02,
"review": 0.98
}
}
]
}Request and response fields follow the API. This example is illustrative, not live inference. The response excerpt omits runtime metadata and timings.
Illustrative workflow. Your application controls the action.Context → understanding → action
The intelligence layer for
whatever comes next.
Emails & conversationsRequests & rulesSearch & retrieved contextYour next idea
Make understanding part of every workflow01 — THE IDEA
Every workflow.
A little more intelligent.
Is this urgent? Is it relevant? Which model? Which team?
Your product makes these calls all day. Siltstone is a specialised model for answering them together, from the same context.
THE SILTSTONE APPROACHRead once.
Read once.
Ask widely.
Ask 1000s of questions at once. The context is read once and reused across them, with answers selected from the choices you define.
See the request ↗your contextRead once
Urgent?↗
Which intent?↗
Within policy?↗
Relevant?↗
Where next?↗
And your next question…+
01 Your context→02 Your questions + choices→03 Structured answers
Built for business. Open to possibility.02 — WHAT YOU CAN BUILD
Put intelligence
in the flow of work.
The text is already there.
Make it useful to your software.
An inbox that knows
02 /⌘
The right model.
For every request.
Check the task against your routing rules. Send straightforward work to a fast model and complex work to a reasoning model.
03 /◇
A checkpoint for
your agent’s next move.
Evaluate proposed actions against the task and your policies. Let your application decide what proceeds and what needs review.
Keep going.
Filter search results. Tag conversations. Qualify requests. Check retrieved context.
Explore the examples ↑Made for the people building what’s next03 — FOR BUILDERS
Your context.
Your choices.
Your next move.
Send text and a set of questions. Get booleans, choices or scores, with model confidence for each answer.
↗
No invented answers outside your choices.You define the valid options. Siltstone selects from them, with no free-text answer to hallucinate.
◌
Confidence you can work with.Set review thresholds and validate them on your own workload.
⌘
You stay in control.Your application decides how each answer becomes an action.
An actual request shapeJSON
POST /ask
{
"document": "We were charged twice. Please get a manager to call me today.",
"questions": [
{
"id": "urgent",
"ask": "Does this need attention today?"
},
{
"id": "team",
"ask": "Which team should handle this?",
"options": [
"billing",
"technical",
"sales"
]
}
]
}↳
One request. Multiple questions.
Ready to connect to your workflow.
The API’s document field accepts your text context: emails, messages, rules and more.
Designed for constrained answers.
Siltstone selects from your defined options. It can still choose incorrectly—use representative evaluations and review thresholds for your application.
Pricing04 — MAKE IT PART OF EVERY WORKFLOW
More answers.
Room to do more.
Low input costs. Free output tokens.
Intelligence you can put to work across your business.
On demand
Early accessStart with the workflow you have in mind.
INPUT
$0.042
per 1M input tokensOUTPUT
$0Free
per 1M output tokens- Ask 1000s of questions at once
- Structured answers with confidence scores
- Your context, your questions, your choices
Indicative input pricing. Final rates and billing terms confirmed during early access.
At your scale
Building Siltstone into a bigger operation?
LET’S TALK ABOUT YOUR WORKLOAD
Make it work
for your business.
Tell us about your volume, latency targets and deployment needs. We’ll work through the right setup and pricing with you.
Talk to the team ↗Early accessWhat would you build
What would you build
if you could
ask more?
Bring a workflow, a set of questions, or your next idea.
Let’s put Siltstone to work.