LoopReply product facts

Configure an Agent through conversation

In Studio, the Architect can change an Agent's instructions, widget appearance and workflow from a conversation with the owner. Settings and Workflow panels show the saved configuration alongside the draft chat preview. Owners can also edit configuration directly.

Instructions, appearance and workflow changes remain drafts until Publish. Knowledge sources, the Agent name, booking links, notification recipients and model-context settings such as response style or language are live settings. Check which field changed before assuming a customer cannot see it. Previewing a workflow with configured external actions can contact those services.

The Versions panel shows draft and published state plus saved Architect changes. Its history is a record of changes, not a set of historical snapshots that can be restored.

Explore Agent configuration or the workflow builder.

Use processed knowledge and saved table data

Agents can use uploaded documents, CSV files, XLSX workbooks and crawled website pages. Attach the source to the intended Agent and wait for processing to complete. A successful upload alone does not mean its content is ready for answers. Crawled pages and imported datasets are saved copies; reprocess or sync them when their source changes.

Supported CSV, XLSX and database datasets retain structured rows. The Agent can filter, sort, count and aggregate complete saved datasets instead of estimating totals from a few retrieved passages. Incomplete row coverage is rejected. Mixed currencies or units must be filtered or grouped before a numeric aggregate. Owners should test representative questions against their source data; an AI-generated answer is not a guarantee of correctness.

Explore knowledge sources and the spreadsheet readiness checklist.

Connect PostgreSQL or MySQL with explicit access

Database sources support PostgreSQL and MySQL. The owner supplies the database endpoint, credentials and a supported SELECT query, tests the connection, previews the result and syncs it into knowledge. Chat answers use the saved result rather than running unrestricted SQL against the database for each message.

Use a dedicated database account with read-only permissions and access only to the tables or views the Agent needs. The connector runs read-only transactions, bounds query time and result size, and rejects private or local network destinations. TLS is enabled by default with certificate verification; a custom trusted CA certificate can be configured when required.

Current import limits

Large or unusually complex sources can reach additional parsing or processing limits. Reduce the input or narrow the database query when the app reports a limit. Subscription allowances for Agents, messages and storage are separate from these technical limits. Review current plans and the allowance shown in your Workspace.

Keep a person in the conversation

Team members with the required access can review conversations in the shared inbox, take over a conversation and reply as a human. Human takeover pauses AI responses for that conversation. The operator can return it to AI or end it. A configured handover workflow step can request human help; it does not guarantee that someone is online or will reply within a particular time.

Explore human handover.

Install the widget and configure integrations

The website chat widget can be embedded on a site where you can add the required script. Connecting an external business system may require API credentials, endpoint configuration, permissions and a workflow action. A website embed or configurable API connection is not the same as a native commerce plugin or an automatically connected messaging channel.

Before launch, test your published Agent on the intended site with your own knowledge, integrations and human support process. For setup questions, contact the team.