Model Hub
Routes each step to a capable model from leading providers. Open to any model.
Technology · Cortex
The right model for every step.
Cortex is the AI layer under your agent. It picks a suitable model for each step, keeps personal data out of prompts, and caches what it can, so answers stay fast and costs stay predictable.
Cortex · one turnmodel hub
- Step
- Understand the question → a fast, low-cost model
- Step
- Plan a refund with policy checks → a more capable model
- Privacy
- Card numbers and personal details redacted before the model
- Cache
- Repeated context reused, not paid for twice
Any modelopen to leading providers, or bring your own
Every turnpersonal data redacted before the model
Same qualityfor every customer, whatever their plan
Cachedrepeated context reused, not paid for twice
What Cortex does
Smart routing, with nothing hidden.
Personal data protection
Personal details, including card numbers, are redacted before a prompt and checked again on the way out.
Caching
Prompt caching reuses context across a conversation, which keeps replies quick and costs down.
Retrieval
Finds the right passages from your Knowledge Hub for each turn.
Summaries
Writes a summary of each resolved conversation for your team and the customer’s profile.
Your keys, your model
On Enterprise, bring your own model and your own encryption keys.
A promise
Same quality for every customer.
Cortex never routes by plan. A small business and a large one get the same model quality for the same work.
- Routine steps use fast, efficient models
- Complex steps escalate to more capable models
- The routing rule is the job, never your price plan
Routing ruleinviolable
- Routes by
- The difficulty of the step
- Never by
- Your plan or what you pay
Inside one turn
What Cortex does in the background.
- 01
Protect
Redacts personal details such as card numbers before anything reaches a model.
- 02
Retrieve
Finds the most relevant passages from your Knowledge Hub.
- 03
Route
Sends each step to a model that fits the job.
- 04
Reuse
Caches repeated context so long conversations stay quick and affordable.
- 05
Summarise
Writes a summary when the conversation is resolved.
Privacy first
Customer data stays yours.
- Personal details are redacted before prompts and checked again on replies
- Card numbers are detected and masked
- Your data is isolated to your workspace
- On Enterprise, bring your own encryption keys
Before the modelredaction
- Customer wrote
- “My card 4111 1111 1111 1111 was charged twice.”
- Model sees
- “My card [CARD] was charged twice.”
- Reply
- Checks the order without ever seeing the card
Why a decision layer
One model for everything vs Cortex.
| One model for every step | Mevrik Cortex | |
|---|---|---|
| Simple questions | Expensive model, slower replies | Fast, efficient model |
| Complex steps | Same model as everything else | Escalates to a more capable model |
| Personal data | Often sent as-is | Redacted first |
| Changing provider | A rebuild | A setting |
Use cases
Where Cortex makes the difference.
Peak season costs
WHENThousands of “where is my order?” chats
AI AGENTRoutes routine steps to efficient models
RESULTFast replies, predictable cost
Card details shared in chat
WHENCustomer pastes a card number
AI AGENTNumber redacted before the model
RESULTHelp given, card never exposed
Your own model
WHENPolicy requires an approved model
AI AGENTBring your own model on Enterprise
RESULTSame agent, your model
Growth or Enterprise
Managed for you, or brought by you.
AX Growth
For growing businesses. Live in minutes.
- Mevrik-managed models through the Model Hub
- Personal data redaction on every turn
- Prompt caching
AX Enterprise adds
Same platform, more depth and control.
- Bring your own model
- Bring your own keys
- Spark CX, Mevrik’s own customer-experience model, on-premise (coming soon)
- Models that run with no external provider, for airgapped deployments
Questions about Cortex.
Which AI models do you use?
Cortex routes between leading models and is open to any model. On Enterprise you can bring your own.
Do cheaper plans get worse AI?
No. Cortex never routes by plan. The same work gets the same model quality for every customer.
Is Spark CX available?
Spark CX, Mevrik’s own customer-experience model, is coming soon, including an on-premise option for Enterprise.
Your AI agent, live in 5 minutes.
Start free, connect your channels, and watch it resolve real conversations today.