Perplexity · Chat model
Sonar for customer support
Yes – it grounds answers in your own content so customers get accurate help, not made-up replies.
Featured on
The model at a glance
The facts, from the source.
Context window
127K tokens
Max reply
8K tokens
Accepts
text
Availability
Proprietary
Verified against the provider.
Where it fits
Sonar across support workflows
How well the model suits each job – grounded in what it can really do, not hype.
Why this matters
What breaks when you run Sonar raw
But real-world support depends more on retrieval, grounding and workflow orchestration than raw model intelligence.
Hallucinates wrong answers. It confidently makes up incorrect details about your product or policies.
Gives stale answers. It repeats outdated info after your docs or policies change.
Lacks customer context. It can’t see the customer’s order or account details to personalize help.
Retrieves inconsistently. It may give different answers to the same question over time.
Drifts off-policy. It may ignore your brand’s voice or rules during long chats.
No human handoff. It can’t easily pass the chat to a real person when needed.
The Chatref way
The model is one layer. Grounding is the rest.
The model is just one layer – grounding, retrieval, and escalation decide how well it works for your business.
If you're deploying AI for customer-facing workflows, the model is only one layer – grounding, retrieval quality, escalation logic and knowledge orchestration usually decide whether it works in production.
Can you use Sonar for customer support?
Yes – it grounds answers in your own content so customers get accurate help, not made-up replies.
What is Sonar's context window?
Sonar can hold up to 127K tokens of context in one conversation.
What inputs does Sonar accept?
Sonar accepts text.
Is Sonar open-weight?
No – Sonar is proprietary and runs through its provider.
Will Sonar make up answers in support?
On its own it can. It confidently makes up incorrect details about your product or policies. A grounding layer keeps every answer tied to your real content.
What does Sonar need to work in customer support?
The model is just one layer – grounding, retrieval, and escalation decide how well it works for your business.
How does Chatref use models like Sonar?
Chatref wraps the model in a grounded layer – it answers from your own content, shows where each answer came from, and hands the chat to your team when needed.