AI agent

AI answers, grounded in your knowledge.

Add your sources, test the answers and hand sensitive or unsupported requests to your team.

See how it works
  • Private beta by invitation. Participation terms are agreed individually with each tester.
Synthetic conversation in doue showing a support response and an explicit transfer to a person.
AI agent knowledge screen with two ready French and English demonstration sources.

Authorized knowledge

You choose what it knows.

Useful policies and content are added as knowledge sources, with their readiness clearly visible.

  • French and English sources are managed separately.
  • Only ready content can support an answer.
  • The team keeps control of the knowledge boundary.

In your workspace

Sol and the copilot have different roles.

Sol replies to customers within its configured framework. The copilot helps team members prepare their work, with human review.

  • Customer conversations and copilot actions have separate allowances.
  • Authorized sources support both uses.
  • Visit the Copilot page to explore the team workflow.
Read the guide

In your workspace

Procedures prepared before activation.

For authorized accounts, your team can define a workflow, review imported proposals, simulate scenarios and have the exact version approved before activation.

  • A procedure complements knowledge; it does not replace it.
  • Available tools limit what each step can execute.
  • Order changes are not promised by this workflow.
Read the guide

Test before automating

Test before you deploy.

The evaluation screen checks the agent against test cases, including those that should trigger a handoff.

  • Expected answers and handoffs are distinguished.
  • Sensitive cases remain in the evaluation set.
  • A visible failure lets the team correct sources or instructions.
AI agent evaluation screen comparing expected answers with cases requiring human handoff.

Capabilities and limits

Capabilities and boundaries.

The agent accelerates repetitive answers without inventing authority over your systems.

Available today

Grounded answers

Connects the response to knowledge available in your account.

Shopify lookup

Looks up order and fulfilment details; proposes Shopify actions that a person approves.

Contextual handoff

Transfers the history, sources, and reason already gathered.

Evaluations

Tests expected answers and handoff cases.

Limits and control

No evidence, no automated answer.

Limits remain visible so the team knows what was automated.

  • Automated answers require sufficient evidence in authorized sources.
  • In Shopify, the agent acts only through the assisted actions you turn on; every refund is approved by a person.
  • Out-of-scope cases are handed off with history and a clear reason.

Try your first questions.

Private beta by invitation. Participation terms are agreed individually with each tester.

Contact us
  • Access by invitation
  • Terms agreed individually
  • Private beta by invitation. Participation terms are agreed individually with each tester.