PepoChat
Also known as: PepoCloud AI Support Agent
PepoChat answers website visitors from a business's own pages, documents, help articles, Notion and Google Drive, and says so when the answer is missing rather than guessing. Thirteen prebuilt integrations and custom HTTPS actions let it check Shopify, WooCommerce and Stripe records, file tickets in help desks and Jira, and alert Slack or Discord during the conversation. Each action declares which response fields the model may read. Escalated chats move to a shared team inbox where staff reply in the same window. The agent books meetings through a PepoSmart scheduling calendar and replies in the language the visitor writes in. Plans are metered by AI replies, starting with a free tier.
PepoChat is an AI support agent that sits on a website and works from content the business supplies: a full site crawl, uploaded PDF and Word files, text snippets and question and answer pairs, published help center articles, and pages synced from Notion and Google Drive. Every reply starts with retrieval over that private knowledge base. When nothing relevant comes back, the agent says it has no information on the question and offers a person instead of composing an answer from general knowledge. Retrieved web content and tool results are wrapped as untrusted data in the prompt, so instructions planted on a page cannot steer it.
Actions come from thirteen prebuilt integrations and from custom actions pointed at any HTTPS API. The agent looks up Shopify and WooCommerce orders, reads Stripe subscriptions and cancels them at period end, creates tickets in Zendesk, Freshdesk, Zoho Desk, HubSpot and Salesforce, opens Jira issues, posts alerts to Slack and Discord, subscribes contacts in Mailchimp and triggers Zaps.
It collects the inputs an action needs from the visitor, chains dependent lookups on its own, and reads only the response fields the operator allows. New actions stay off until a test run succeeds. A workspace holds up to 20 actions, and because execution is at least once, customers are told not to connect refunds or payments.
Handoff is built into the conversation model. A chat is unresolved while the agent answers, escalated once a person should take over, and resolved when it closes. The agent escalates when a visitor asks for a person, when it cannot find an answer, or when a tool or the model fails, and staff hand a chat back by returning it to unresolved. Business hours and an email follow up mode change what the agent promises when nobody is watching the inbox. Operator instructions shape tone and add refusals, but they sit beneath base rules that keep the agent grounded and keep handoff available.
Meeting booking and meeting memory both run through PepoSmart, a sibling scheduling and meeting notes product. With it connected, the agent offers real openings in the visitor's timezone and books them, and for visitors who verify an email with a one time code it can recall summaries and commitments from earlier recorded meetings. Voice is a separate channel run on the customer's own Vapi account, without the chat agent's knowledge or tools. The service is hosted by PepoChat on OpenAI models, with no model choice, no stated hosting region and no server API beyond a small widget control script.
Vendor details
Canonical URL
https://pepochat.com
Category
Customer support agent
Subcategory
Website support agent grounded in owned content
Funding status
Operated by PepoCloud LLC, Lewes, Delaware. No funding disclosed.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Prebuilt: Shopify, WooCommerce, Stripe, Zendesk, Freshdesk, Zoho Desk, HubSpot, Salesforce, Jira, Slack, Discord, Mailchimp and Zapier. Custom HTTPS actions with GET, POST, PUT, PATCH or DELETE and encrypted headers. Knowledge from Notion and Google Drive. Booking and meeting memory through PepoSmart; voice through the customer's own Vapi account. Installs by script tag, WordPress plugin, React or Next.js snippet, or a hosted chat page.
In practice
A shopper asks where an order is; the agent requests the order number, reads the live status from Shopify and replies with the shipping stage without opening a ticket.
A subscriber asks to cancel; the agent finds the Stripe customer by email, reports the renewal date and sets cancellation at period end only after the customer confirms.
A visitor reports a checkout outage; the agent posts an alert with a summary to a Slack channel, escalates the chat, and a teammate replies in the same window.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
6.5 / 14 capabilities · 46%
| Integrations & Tool Calling | Full |
|---|---|
|
Thirteen prebuilt integrations create ready actions: order lookups in Shopify and WooCommerce, subscription reads and cancellation at period end in Stripe, ticket creation and status checks in Zendesk, Freshdesk, Zoho Desk and Salesforce, contact lookups and tickets in HubSpot, Jira issues, Slack and Discord alerts, Mailchimp subscriptions and Zap triggers. Custom actions call any HTTPS endpoint with GET, POST, PUT, PATCH or DELETE, up to ten collected inputs and encrypted headers. Each action lists the response fields the model may read, and a Test control runs the real request before the action is enabled. A workspace holds up to 20 actions, and execution is at least once, so a rare retry can repeat a write. Sourcepepochat.com/docs/integrations/overviewread 2026-10-26 |
|
| Workflow Orchestration | Partial |
|
At run time the agent picks the enabled action that fits a request from its when to use text, collects missing inputs from the visitor and chains dependent calls itself, as when an email lookup feeds a Stripe customer search and then a subscription check. Conversation statuses route a chat between the agent and the team. There is no builder for customer defined sequences, branches, conditions or retries; the order of steps comes from the model and the action descriptions rather than an editable flow. Sourcepepochat.com/docs/integrations/striperead 2026-10-26 |
|
| Knowledge Grounding & RAG | Full |
|
Knowledge comes from a whole site crawl counted as one source, uploaded PDF and Word files, text snippets and question and answer pairs, published help center articles, and pages synced from Notion and Google Drive, limited to the files and pages the operator shares. Each reply retrieves the most relevant passages from that private store and writes from them; the agent does not browse the open web or answer from general knowledge about the business. When retrieval finds nothing, it says so and offers a person. Retrieved content is wrapped as untrusted data, so text planted in a page cannot redirect the agent. Sourcepepochat.com/docs/knowledge-base/how-answers-workread 2026-10-26 |
|
| Human Oversight & Guardrails | Partial |
|
Escalation fires on a request for a person, an accepted offer after a failed search, a tool or model failure, and an exhausted free allowance, and an escalated chat stops the agent until staff set it back to unresolved. Destructive steps wait for the visitor: Stripe cancellation needs an explicit request and a confirmation, and only the visitor's own request can trigger a booking or an action. Operator instructions add refusals beneath base rules that keep grounding and handoff in force. Staff do not approve individual actions before they run; tickets, alerts and lookups execute as soon as the agent calls them. Sourcepepochat.com/docs/conversations/human-handoffread 2026-10-26 |
|
| Security, Identity & Governance | Partial |
|
Integration keys, OAuth secrets, refresh tokens and custom header values are encrypted with AES-256-GCM and never shown again, and each workspace keeps its knowledge, conversations, contacts and credentials in separate namespaces. Members sign in with a verified email and password or with Google, with optional authenticator app codes and backup codes for two factor sign in. Roles are owner, admin and member, and members can still manage integrations and billing; only owners and admins set retention. Outbound calls are HTTPS only, with private hosts blocked, redirects refused and responses capped. There is no single sign on or administrative audit log, and the company states it does not claim SOC 2, ISO 27001, a hosting region or a signed DPA template. Sourcepepochat.com/docs/security/privacyread 2026-10-26 |
|
| Observability & Auditability | Partial |
|
The team inbox shows each conversation live with its status, assignee, internal notes and the visitor's device, language and session details, and a CSV export carries transcripts labeled by visitor, agent and operator for a date range. Analytics adds satisfaction votes from resolved chats and AI labeled topics with weekly trends. Meeting memory reads are logged with the workspace and conversation. The record covers what was said; tool requests and responses do not appear as steps, and there are no charts of volume, resolution rate or reply usage. Sourcepepochat.com/docs/conversations/team-inboxread 2026-10-26 |
|
| Memory & State Persistence | Partial |
|
A visitor session lasts 24 hours and extends while the visitor chats, so the conversation and the Messages tab carry over in that browser, and every conversation stays in the team inbox. For visitors who verify an email with a one time code, meeting memory retrieves dates, interests, commitments and summaries from earlier meetings recorded in PepoSmart, never raw transcripts, and only when the workspace switches it on. No separate memory of earlier chat conversations is described; recall beyond the session depends on that PepoSmart connection. Sourcepepochat.com/docs/conversations/meeting-memoryread 2026-10-26 |
|
| Deployment & Data Residency | Not documented |
|
The service runs on a managed stack, with the backend on Convex, the dashboard and widget on Vercel and replies generated by OpenAI models. No hosting region is claimed, and there is no self hosted, private cloud or regional option. A script tag, WordPress plugin, React or Next.js snippet and hosted chat page change where visitors reach the agent, not where it runs. Allowed domains and blocked countries restrict where the widget may start a session. Sourcepepochat.com/docs/security/privacyread 2026-10-26 |
|
| Prebuilt Agents / Templates / Packs | Partial |
|
Each prebuilt integration card walks through connecting the tool and adds ready actions such as order status, subscription status or ticket creation, with input formats already validated. A WordPress plugin installs the widget without code, and example instruction lines can be added with a click. The product ships one support agent configured from the customer's content; there are no named role agents, industry packs or reusable workflow templates beyond those action cards. Sourcepepochat.com/docs/integrations/overviewread 2026-10-26 |
|
| Triggers & Channel Coverage | Partial |
|
The agent wakes on a visitor message in the website widget or the hosted chat page and answers around the clock, with business hours shaping what it promises at handoff. A voice call and a phone number can appear in the widget, but voice runs on the customer's own Vapi assistant without the chat agent's knowledge or tools, and calls are not saved to the inbox. New lead notifications can email the team, post to Slack or call a webhook. There is no email, WhatsApp, social or messaging app channel, and no scheduled or inbound webhook trigger. Sourcepepochat.com/docs/widget/voiceread 2026-10-26 |
|
| Model Flexibility & Routing | Not documented |
|
Replies and topic labels use OpenAI models, and document extraction uses a vision capable model for scanned pages and images. The customer does not choose a provider or model, bring its own key or route tasks between models; tone and behavior are shaped through instructions instead. Sourcepepochat.com/docs/knowledge-base/how-answers-workread 2026-10-26 |
|
| APIs / SDKs / MCP Extensibility | Partial |
|
Custom actions let the agent call a company's own HTTPS services with templated paths, bodies and encrypted headers. On the page, a four method script interface opens, hides, removes and reinitializes the widget, and the WordPress plugin offers filters that control where it renders. New leads can be pushed to a webhook. There is no REST or server API to send messages, manage knowledge or pull conversations, no SDK beyond install snippets, and no MCP server or client. Sourcepepochat.com/docs/widget/behavior-and-apiread 2026-10-26 |
|
| Testing, Debugging & Optimization | Partial |
|
A Test the agent drawer runs the live widget with free replies kept out of analytics, while bookings and actions still execute for real. Every action card has a Test control that sends the real request with sample inputs and shows the status and response before the action is switched on. After launch, satisfaction votes and rising topics point to gaps in the content. There are no saved test sets, automatic scoring, model comparisons or regression checks. Sourcepepochat.com/docs/widget/customizationread 2026-10-26 |
|
| Browser / Computer-use | Not documented |
|
The agent works through retrieval and API actions. A site crawl reads pages into the knowledge base, and visitors can attach images that the agent reads, but nothing operates a browser or desktop on the agent's behalf, and work without an API path is handed to staff. Sourcepepochat.com/docs/knowledge-base/how-answers-workread 2026-10-26 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
$49 per month
Monthly AI reply allowance plus seats
Included quota
Starter: 2,000 AI replies a month, unlimited sources up to 200 MB of text, 300 help center articles and three members.
What is public
Free $0, Starter $49 a month and Growth $149 a month, with reply allowances, seat counts, knowledge limits, overage and seat prices all listed.
Billing mechanics
Each AI reply counts against the monthly allowance; test chats do not. Paid plans bill overage on the next invoice, and Free hands extra conversations to the inbox.
Cost watchouts
Voice uses the customer's own Vapi account, billed by Vapi. Appointment booking and meeting memory require a separate PepoSmart account whose pricing is not on this page. A workspace holds up to 20 actions on every plan.
Variable cost rationale
Reply volume past the allowance bills at a flat published rate and seats add predictably, but voice runs on a separately billed Vapi account and booking and meeting memory need a PepoSmart account.
Additional watchouts
Price voice on Vapi and booking on PepoSmart before comparing totals with suites that bundle them.
Overage / add-ons
Replies past a paid allowance cost $10 per 1,000 on the next invoice. On Free, replies past 200 hand conversations to the team inbox instead of billing. Extra seats cost $15 a month, or $150 a year on yearly plans.
Sales call required
No, self serve available
Free / trial
Free plan with 200 AI replies a month, five knowledge sources and one member; no card required
Lowest paid plan
Starter
Commercial notes
Self serve with no card on Free. Yearly billing gives two months free on paid plans. No enterprise tier is listed.
Key ambiguities
The plan page says every feature is on every plan, while the comparison table reserves new lead emails, branding removal, priority support and yearly billing for paid plans; the free plan summary also describes knowledge sources as a monthly allowance, though they are a standing limit.
Cancellation / refund
Monthly plans run month to month; yearly plans are billed up front at ten months. Both prorate when moving between plans or intervals, and a downgrade returns to Free without losing conversation history.
Support SLA / resale
Priority support on Growth; no SLA published.
Missing data
PepoSmart pricing, refund terms and any volume or enterprise arrangement.
Related vendors
- Decagon — AI agent platform purpose-built for customer support, with deep…
- 5.Y — 5
- Ada — Agentic customer experience platform whose AI agents autonomously…
- Aisera — Enterprise agentic AI platform that orchestrates specialized agents…
- ASAPP — Generative AI for enterprise contact centers that plugs into…
- Assembled — AI customer support orchestration platform that unifies autonomous…