Support that runs itself
Stop answering the same questions over and over. Let AI handle the repetitive stuff so your team can focus on what matters.
Supported answers
When evidence is insufficient
Workflows to maintain
Automation that actually works
Powered by AI, not complex rule builders
AI auto-responses
AI answers supported common questions from your documentation and shows the sources behind the result.
Independent of support hours
Published help content remains available for visitors to search while your team is offline, subject to service availability.
Uses the content you update
When you update and publish source content, later retrieval uses the updated material. There is no separate model-training workflow.
Verified handoff
An unsupported answer can offer a visitor-confirmed handoff with bounded context, but only after your Telegram destination passes a real delivery test.
Proactive suggestions
AI suggests relevant articles as customers browse. Help before they even ask.
Direct setup path
No complex automation rules to configure. Connect your docs and AI handles the rest.
What can be automated?
Supported common questions can be answered from published sources
No complex setup required
Connect docs
Crawl published pages, connect GitHub, run the CLI, or import help articles from Intercom, Zendesk, and Freshdesk.
Review sources
Review the imported or generated material and publish only the content the agent may use.
Automation works
Supported questions receive answers from the published sources, with citations.
Frequently asked questions
Routine questions with clear published evidence are the best candidates: how-to guides, feature explanations, pricing facts, and basic troubleshooting. Account-specific actions, exceptions, and unsupported claims stay on the human path.
You do not need to build a decision tree. You do need to add and publish reliable sources, configure any claims-policy topics, and verify the human path you want to offer. Re-import or rescan sources when they change.
HelpShelf withholds the unsupported answer. When your human path is verified, the visitor can choose to send a bounded summary and reply address to your team. Internal interaction records may still be created, but the model does not alert a human or claim the handoff was sent without visitor confirmation.
HelpShelf verifies material claims against retrieved passages from your published documentation. Confident answers show their sources. When an answer cannot be verified, HelpShelf says so and offers a visitor-confirmed handoff only when a tested destination is reachable.
HelpShelf uses a conversational response style, but complete answers still require support from published sources and inline citations. Unsupported widget responses stop clearly; a human handoff is visitor-confirmed and available only through tested Telegram.