Rules first, model second
Each incoming email is compared against your rules for sender, subject and body. Matches are handled immediately with no model call. Only unmatched mail is passed to the LLM agent, which classifies it and chooses an action from the tools you enabled. This keeps behaviour predictable and costs low.
What the agent can do
- Classify intent such as enquiry, complaint, invoice, booking or spam.
- Reply, forward, archive or escalate according to your settings.
- Look up context through CRM, ticketing, calendar or finance tools.
- Skip messages it is unsure about and label them for a person to review.
Getting started
- Open the ERA workflow template and connect your mailbox.
- Add a few rules for your most common message types.
- Pick the tools and language-model preferences for the agent.
- Review the summary and enable the workflow.
Governance and visibility
Executions are listed with their status, so you can see when a run started, whether it succeeded and what it processed. Labels in the mailbox show outcomes at a glance. You can refine rules whenever you see a pattern the agent keeps handling.
Frequently asked questions
What if the AI gets a message wrong?
Uncertain messages can be skipped and labelled for human review, and you can add a rule so the same pattern is handled predictably next time.
Is my email content used to train models?
Coderva sends message content to the language model configured for your workflow to make routing decisions. Review the privacy policy and your model preferences for details.
Is this different from a regular rule engine?
Yes. A rule engine only handles what you anticipated. The hybrid design adds a model for the unexpected cases while keeping the rules you rely on.