An AI agent that answers from your own docs.
Cove AI reads each app's help center, website, files, and Q&A, then answers customer questions directly, in their language, with citations. When it isn't sure, it hands off to your team with the full context. Not a generic chatbot: an agent built to deflect tickets without guessing under your name.
Grounded in your knowledge · Escalates when unsure · $0.20 per resolved conversation
It answers from your knowledge, and only your knowledge.
Cove AI is trained on your own help-center articles, a crawled docs or help site, uploaded files, and hand-written Q&A, kept separately for each app. Every reply is grounded in what it retrieved: hybrid search plus a reranker pick the right passages, and Cove AI cites them back to the customer. If the answer isn't in your sources, it is built to hand off instead of guessing a price, a policy, or a step.
- ✓Four sources: help center, website crawl, uploaded files (TXT/MD/CSV), Q&A
- ✓Hybrid retrieval (semantic + keyword) with a cross-encoder reranker
- ✓One-click "Train AI", embeds your existing articles automatically
- ✓Knowledge kept per app, so each app's customers get that app's answers
- ✓Source citations shown to the customer, so every answer can be checked
- ✓Follow-up aware, resolves "what about mine?" into a full question before it answers
Start in Suggest. Graduate to Auto when it's earned your trust.
Cove AI starts safe: it drafts replies your agents approve. As your team accepts those drafts, the analytics show the acceptance rate climbing, and tell you when Cove AI is ready to reply on its own. Flip it to Auto, or Auto-only-when-offline so Cove AI covers nights and weekends while your team owns the day. Hybrid does both automatically.
- ✓Five modes: Off · Suggest · Auto-when-offline · Auto · Hybrid
- ✓Suggest mode drafts replies; agents accept, edit, or reject in one click
- ✓Trust funnel surfaces acceptance rate and a "ready for Auto" signal
- ✓Off-hours coverage tied to your live availability + business hours
- ✓Handles greetings and small talk itself, never pages a human for "thanks"
Cove AI knows your plans. For Shopify apps, it knows who pays.
Describe your tiers once and Cove AI answers plan questions precisely, citing only the prices and links in your own sources. Connect your Shopify Partner account and it also sees each merchant's billing: free install or paying, MRR, and Shopify Plus. So when a free-install merchant asks about a paid feature, Cove AI is honest that it's paid instead of pretending it doesn't exist, and can nudge an upgrade when you want it to.
- ✓Shopify apps: real billing facts, paying vs free install, MRR, Shopify Plus
- ✓Honest about paid features, never a blank "not available"
- ✓Neutral or nudge-to-upgrade, your call, per app
- ✓Describe your tiers once; Cove AI answers plan questions precisely
Built to know its limits, and hand off cleanly.
A confidence threshold decides when Cove AI answers versus escalates. Set topics it must always send to a human (billing disputes, cancellations, legal) and Cove AI routes them straight to your team. When it does hand off, it writes a warm, context-specific line so the customer gets a real transfer, not a cold dead end. Customer messages are treated as untrusted, so prompt-injection attempts get politely declined.
- ✓Tunable confidence threshold: answer more, or only when very sure
- ✓Scope fence: topics Cove AI always routes to a human
- ✓Max replies before handoff, and a clear AI disclosure label
- ✓Warm, context-aware handoff line on every escalation
- ✓Prompt-injection resistant, customer input is never trusted as instructions
See exactly what Cove AI deflects, and what it's missing.
A live dashboard shows deflection rate, conversations handled, CSAT, and cost over time, answered versus escalated, day by day. And Knowledge Gaps clusters the recurring questions Cove AI couldn't answer, so you know precisely which Q&A or article to add next to deflect more. You're never guessing, and at $0.20 per resolved conversation, the cost is always clear.
- ✓Deflection rate, handled count, CSAT, and cost, by date range
- ✓Answered-vs-escalated trend with deflection % overlaid
- ✓Knowledge Gaps: recurring unanswered questions, clustered and counted
- ✓Haiku by default for low cost; Sonnet when you want higher quality
Most AI chatbots would rather guess than say "I'm not sure." Cove AI hands off.
Generic support bots optimize for always having an answer. For a software product, a wrong answer about billing or a setting is worse than no answer: it creates a ticket and a bad review. Cove AI is built the other way: grounded in your docs, honest about your plans, and quick to hand off when it isn't sure. It deflects the easy questions so your team spends its time on the customers who actually need a human.
How does Cove AI compare with Intercom Fin and other AI support agents?
Five of the ten AI support agents we checked say in their own documentation that they answer only from your content (Cove AI, Plain's Ari, Help Scout, Gorgias and Tidio), and three tie hand-off to the agent's own uncertainty (Cove AI, Plain's Ari and Gorgias). Only Cove AI and Plain's Ari do both and also show sources to the customer. Cove AI's $0.20 per resolved conversation is the lowest published dollar price of the six; Plain prices in credits, and Intercom's Fin costs from $0.99 per outcome. Every row below comes from each vendor's own documentation, such as Plain's Ari docs and Intercom's pricing page, checked in October 2026.
| AI agent | Answers only from your content | Shows sources to the customer | Hands off when | AI price |
|---|---|---|---|---|
| Convot Cove AI | Yes | Yes | Below a confidence threshold you set, fenced topics, reply cap | $0.20 per resolved conversation |
| Plain Ari | Yes | Yes | The moment it is unsure | 100 credits per resolved thread (no dollar rate published) |
| Help Scout AI Answers | Yes | Yes | Customer asks, guardrail topics | $0.75 per resolution |
| Gorgias AI Agent | Yes | Not stated | Anger, a human request, listed topics, low confidence | $0.85 to $1.00 per automated interaction |
| Intercom Fin | Uses your sources; "only" not stated | Yes | Customer asks, frustration, high-risk content, your rules | From $0.99 per outcome |
| Zendesk AI agents | Uses your sources; "only" not stated | Yes | Complex, urgent or sensitive, via escalation flows | Rate not published |
Convot publishes this page and sells Cove AI. Documented controls are not measured accuracy, so test any agent on your hardest questions in suggest mode first. The table shows six of the ten; Tidio, Freshdesk, Crisp and Front are in AI support agents that cite their sources and hand off.
Cove AI questions
What is Cove AI?
Cove AI is the AI support agent built into Convot, a help desk for small software teams. It answers customers from each app's own help center, web pages, files and hand-written answers, can show the source of each answer, and hands the conversation to a person when it is not confident or the topic is on a list you set. It costs $0.20 per resolved conversation on Convot's paid plans.
How much does Cove AI cost compared to Intercom Fin?
Cove AI costs $0.20 per resolved conversation in which it sent a reply, charged once per conversation. Intercom lists Fin from $0.99 per outcome, so 2,000 resolutions a month cost $400 on Cove and at least $1,980 on Fin. The two count a resolution differently, so compare definitions as well as prices.
Does Cove AI only answer from my own docs?
Cove AI answers only from four knowledge sources you connect for each app: its help center, crawled pages, uploaded files and hand-written answers. Separately, it can read live account data, such as a customer's plan, from your own REST endpoint or MCP server. When nothing matches well enough, it says so and hands off instead of guessing.
When does Cove AI hand off to a person?
Cove AI hands off when its confidence is below the threshold you set, when the topic is on your list of topics that always go to a person, such as billing disputes, or when it reaches the cap on bot replies in one conversation. The person who picks it up sees the whole conversation, so the customer does not repeat themselves.
Which AI support agents hand off to a human when they are unsure?
Cove AI, Plain's Ari and Gorgias document hand-off tied to the agent's own uncertainty: Cove AI below a confidence threshold you set, Plain "the moment it is unsure", and Gorgias when it "doesn't have enough knowledge to answer reliably". Tidio and Crisp hand off when the agent cannot answer, and Intercom, Zendesk and Help Scout mainly when the customer asks or a topic you marked comes up.
Can I review Cove AI's answers before customers see them?
Cove AI starts in suggest mode, where it drafts replies and your team reviews, edits or rejects them before anything is sent. You can then turn on automatic replies outside working hours only, or at all times, once the drafts hold up.
How do I stop an AI support bot making things up?
Stop an AI support bot making things up by limiting it to approved sources, showing a source with every answer, setting a confidence threshold that forces a hand-off, fencing off topics such as billing disputes, and testing it in suggest mode before customers see it. Cove AI has each of these controls built in.
Cove AI works tighter with the rest of Convot.
AI escalation & QA
Cove AI answers the front line. Behind it, frustration escalation and auto-QA grading keep every conversation in check.
See the AI layer →Help center
The articles on your own domain are exactly what Cove AI trains on, write once, deflect twice.
Explore help center →Live chat
Cove AI lives in the branded widget on each of your apps, with a clean human handoff when needed.
Explore live chat →Put Cove AI on your front line.
Start Convot free, then add Cove on a paid plan. Train it on your docs in one click, run it in Suggest mode, and pay $0.20 per resolved conversation in which it replied.