One platform. Every surface.
The embeddable widget, the agent inbox, the knowledge graph, and the customer-facing help center — all sharing the same data. AI-first by default. Agentic when connected.
Answers in seconds
Real RAG with citations. The model can't answer from training data — only from your docs.
Live in five minutes
One script tag or one npm install. No migration project, no kickoff call, no consultant.
Smarter every day
Each closed conversation feeds the knowledge graph. Approve a fix once, the AI uses it forever.
Resolve. Don't just deflect.
Most support AI is a glorified FAQ search. Ours runs hybrid retrieval over your docs and past tickets, cites the source on every reply, and refuses to hallucinate policy. If retrieval comes up empty, it falls back to the script you wrote — never the model's training data.
Real RAG. Real citations. Zero made-up answers.
Hybrid pgvector + Postgres full-text retrieval, optional Cohere reranking, and a pre-retrieval rules layer for the answers you want the AI to give verbatim. Strict context-faithfulness instructions keep the model honest.
Grounded by default
Every reply cites the source. No source, no answer — the model uses your fallback copy verbatim.
Rules layer over retrieval
Canned responses, term aliases, and admin overrides resolve before retrieval. You stay in control of what the AI says.
Streaming, multi-model
Groq Kimi K2 by default for speed. Auto-swaps to Claude Sonnet 4.5 for tool calls. OpenAI optional. Per-chatbot choice.
Don't just answer it. Solve it.
Connect Shopify in two clicks. Mime Chat's AI runs a multi-step tool loop on Claude Sonnet 4.5 — it cancels orders, issues refunds, and updates shipping addresses for every customer who asks, automatically. You only pay when it actually does the action.
The tool loop, made visible.
Read tools verify. Write tools resolve. Every tool call is logged in the conversation timeline, attached to the assistant reply, and counted as a billable resolution only when the action completes successfully.
When AI hands off, your team picks up where it left off.
Not a sidekick to Help Scout. A complete replacement. Live inbox, presence, internal notes, saved replies, round-robin assignment, real-time typing — the full agent surface, with the AI's transcript and reasoning already in the thread.
Full transcript transferred
Agent inherits the AI's reasoning trail — retrieval chunks, citations, confidence scores. The customer never repeats themselves.
Presence & round-robin
Online / away / offline status. Auto-assignment to the agent with the fewest active threads. Toggleable per org.
Notes, replies, attachments
Internal notes invisible to visitors. Saved replies. File and image attachments with inline preview. Tags for triage.
CSAT, built in
Visitor thumbs-up/down + optional text after every close. Persisted, surfaced, and counted in your weekly digest.
Pickup from anywhere.
Native iOS and Android apps for agents. Push notifications for new conversations and AI escalations. Deep-link from notification straight into the thread. Passwordless sign-in.
It gets smarter every day.
Most "AI gets smarter" claims are vapor. Ours is a real, persisted workflow. Conversations close. Operators triage. The next customer with the same question gets the curated answer instantly.
Conversation closes.
AI-resolved, human-resolved, or stale-swept. Mime Chat creates an Improve candidate with retrieval, escalation, and citation diagnostics — and recommends a fix.
Your team approves with one click.
Triage in the Improve workspace. Queued → approved → published. Edit the answer if you want; ship the AI's draft if you don't.
The AI uses it on the next customer.
Publishing creates or updates a knowledge source. Re-chunked, re-indexed, re-embedded — live in retrieval within seconds. Next time, the answer is instant.
The full support stack. None of the bloat.
You shouldn't need a chatbot, a helpdesk, a help center builder, and a forms tool stitched together with webhooks. Mime Chat is one platform — the embeddable widget, the agent inbox, the knowledge graph, and the customer-facing help center, all sharing the same data.
AI-first conversations
Hybrid retrieval, citations, multi-model. Pick Groq for speed, Claude for tools, OpenAI for embeddings — per chatbot.
Real RAG with citations
pgvector + full-text merged by chunk id. Optional Cohere reranking. Pre-retrieval rules layer.
Script tag or React
One-line embed for any site. @mimechat/embed-react for full programmatic control in modern apps.
Built-in help center
Categories, articles, themes. Public articles double as RAG knowledge — one source of truth.
Forms builder
Drag-and-drop. Field library, live preview, readiness bar. Lead capture, surveys, structured intake.
Conversation insights
First-response time, deflection, agent performance. Cross-conversation patterns: "37 asked about refunds; you don't have an article."
Business hours that change how the AI talks.
Per-chatbot weekly schedule, holiday overrides. Out-of-hours, the AI stops promising live handoff and switches to email follow-up copy.
Native iOS & Android
Real agent app. Push for new conversations & AI escalations. Deep-link from notification to thread.
Visitor identify API
Tell the AI who's asking. Plan, lifetime value, custom attributes — all injected into the system prompt.
Capture visitors naturally
No pre-chat forms. The AI asks for name and email in 1–2 short sentences. Validated, with intent shortcuts.
MCP server for content
Edit your help center from Cursor, Claude Code, or any MCP client. @mimechat/mcp-server on npm.
Agentic ecommerce actions
Cancel, refund, update shipping — from the chat. $0.99 per resolution, only when the action completes.
Install in two minutes. Migrate over the weekend.
One script tag for vanilla sites. One npm install for React apps. Tell the AI who's asking with a single function call. No migration project, no implementation consultant, no kickoff Zoom.
<!-- in <head> --> <script src="https://cdn.mime.chat/embed.js" data-embed-key="emb_8a3f…" defer></script>
import { MimeChatWidget } from "@mimechat/embed-react"; export default function Layout({ children }) { return ( <> {children} <MimeChatWidget embedKey="emb_8a3f…" baseUrl="https://app.mime.chat" /> </> ); }
window.mimeChat("identify", { email: user.email, name: user.name, customAttributes: { plan: "Pro", lifetimeValue: 1240, accountAge: "2y", } });
Stop scaling headcount.
Start scaling knowledge.
The companies that win at support in the next decade won't be the ones with the biggest queues. They'll be the ones whose answers compound. Mime Chat is built for them.
Start free. Scale when you're ready.
Per-conversation pricing — you pay for outcomes, not seats. Every paid plan includes a 7-day free trial. Annual saves up to 40%.
Billed monthly. 7-day free trial included on every plan.
For teams getting their first widget out the door. Everything you need to launch AI-powered support.
- 1 chatbot
- 2 agent seats
- 500 AI conversations / month
- Basic support center (FAQs)
- Standard widget embed
- Email support
The plan most teams run on. Handoff, insights, and the knowledge pipeline that makes everything better.
- 5 chatbots
- 5 agent seats
- 3,000 AI conversations / month
- Live human handoff
- Conversation insights
- Knowledge pipeline (chat → article)
- Custom widget branding
For organizations with compliance, residency, and audit requirements — and the teams that need them.
- 20 chatbots
- 15 agent seats
- 15,000 AI conversations / month
- Advanced analytics & reporting
- Custom domains & widget options
- Dedicated onboarding session
- Priority support SLA
For teams with security & compliance requirements. SSO, HIPAA, custom roles, dedicated CSM.
- Unlimited chatbots & agent seats
- SSO, HIPAA, SLAs
- Custom roles & permissions
- Dedicated customer success manager
- Volume discounts on AI usage
Self-serve everything. Upgrade, downgrade, or cancel in two clicks — no email-our-CSM theatre. Live usage and limits in your dashboard.
Domain-locked embeds
Whitelist origins per widget. No unauthorized embeds.
Distributed rate limiting
Sliding-window via Upstash Redis. Built for spikes.
Encrypted integrations
AES-GCM at rest for Shopify and OAuth tokens.
Sentry-monitored uptime
Errors and replays across the dashboard. Embeds get zero replay to protect end-user PII.