Bot to human, human to bot
The AI handles repeatable questions. The moment confidence drops or the visitor asks for a person, it hands off with full transcript and intent. Agents can also hand it back when they're done.
Live · v0.1 · Bring your own LLM
Talk to your customers.
Powered by the AI you already trust.
OpenAgent is a live chat and AI support platform that plugs into your Gemini, OpenAI, Claude, OpenRouter or self-hosted model. Your keys, your data, your costs. From $3 per site, per month billed annually ($36/year), or $5 billed monthly.
No card required. Paste a model key and your widget is live in about 10 minutes.
- Unlimited live chat
- RAG knowledge base
- Embeddable widget
- GDPR-ready, your data
Want a live demo? The chat bubble in the corner is OpenAgent, answering from our own docs.
A
Aria · AI agent
Powered by your Gemini key
Keys never leave your stackyour Gemini key
Message Aria…
Built for teams shipping fast across
E-commerce support
SaaS onboarding
Edtech help centres
Healthcare triage
Fintech KYC chat
Hospitality concierge
B2B sales rooms
Logistics ops
E-commerce support
SaaS onboarding
Edtech help centres
Healthcare triage
Fintech KYC chat
Hospitality concierge
B2B sales rooms
Logistics ops
See it work
Five moments where OpenAgent earns its keep
Not a feature list. These are the situations your visitors put you in every day, and what handling them looks like.
Around the clock
It's 2:47 AM. Your visitor still gets a real answer.
Most questions arrive when nobody is at a desk. The assistant answers from your knowledge base in seconds, any hour, any timezone, and your team finds a clean transcript instead of an overnight queue.
- Answers in seconds, 24/7
- Full transcript for the morning shift
- Works in your visitor's language
Acme Support · onlineWidget · acme.com
Tonight · 2:47 AM
Do you ship to Ireland? Need it before Friday.
Acme AssistantAI
Yes. Irish orders ship from our Dublin hub, 2 to 3 working days, so ordering today lands it by Thursday. shipping-policy.pdf · section 2
Answered in 4 seconds. Nobody woken up.
Live visitors
Someone's been on /pricing for four minutes. Go say hi.
See who is on your site right now, which page they are reading, and how long they have been there. Open the chat yourself when a visitor lingers on a page that signals buying intent, and let Slack tell the team the moment they write in.
- Live page + journey per visitor
- Start the conversation proactively
- Message alerts in Slack, Google Chat, push
Live now · 3 visitorsDashboard · Visitors
P
PriyaIndia
/pricing· 4m 12sHigh intent
D
DanielUnited States
/compare/vs-intercom· 1m 40sHigh intent
L
LenaGermany
/guides/how-rag-works· 0m 55s
S#sales · Priya has been on /pricing for 4 minutesnow
Human + AI, both directions
AI answers what it knows. People handle what it can't.
When the visitor asks for a person, or the question needs judgment, the conversation escalates with the full transcript, an SLA timer and a Slack ping. When the human is done, one click hands it back and the AI resumes the night shift.
- Escalation with SLA timers and routing
- Agent and AI can share one conversation
- Hand back to AI when you're done
Conversation #4821 · PriyaInbox
Acme AssistantAI
Your invoice history is under Billing → Statements.
I was charged twice. Can I talk to a person?
Escalated to Maya · SLA timer 2 min · Slack pinged
Maya · Billing
So sorry about that, Priya. I can see the duplicate charge and I have just refunded it. You'll see it back in 2 to 3 days.
Maya handed the conversation back to AI
Acme AssistantAI
I'm back on watch. Anything else you need tonight?
Grounded answers
Trained on your docs. Cites its sources. Never bluffs.
Upload PDFs, paste URLs, drop in text. Every reply is retrieved from that knowledge base and linked back to the exact source. When the answer is not there, the assistant says so and brings in a human instead of inventing one.
- Citations on every answer
- PDFs, URLs and raw text in minutes
- Honest handoff instead of a guess
Knowledge-grounded replyRAG pipeline
What's your refund window?
Searching your docsrefund-policy.pdf · ¶2terms.md
Acme AssistantAI
Full refund within 30 days of purchase, no questions asked[1]. After 30 days we offer account credit[2].
And when it's not in your docs
Visitor: Do you integrate with SAP?
Assistant: That's not in my knowledge base, so I won't guess. Let me bring in the team.
Identify & capture
Ask for exactly what your pipeline needs. It lands in the CRM by itself.
Choose what the widget asks for: name, OTP-verified email, validated phone, or your own custom fields like order ID or company size. Every answer is written to the visitor's contact record, ready to export or push over the API.
- Per-site configurable fields
- Email OTP and phone validation built in
- Flows straight into the built-in CRM
Before we start, a few detailsWidget · identify step
Name
Priya Nair
Work email
Phone
+91 98765 43210
Order ID
#A-1042
Saved to the contact record · tagged “pricing-page”
What's inside
Everything you'd expect from a real support stack, without the enterprise price tag.
Live chat, AI, knowledge, CRM, alerts, routing. One platform, one login, your data. No per-seat upcharges, no AI surprise bills.
Multiplayer AI
Agents and the AI work the same conversation together, live. See which teammates have a chat open, watch typing as it happens, and pull the AI in on demand: type @ai or hit Ask AI for a draft you can edit and send, or post as the AI in one click. No more silent bot the moment a human steps in.
How Multiplayer AI works →Instant visitor alerts
Browser push, in-app pings, sound, email, Slack and Google Chat. Pick your channel and a visitor message lands there the second it's sent. Per-agent quiet hours and routing rules so the right person sees the right chat.
Chat with live visitors
See who is on your site right now, what page they're reading, where they came from. Start a conversation proactively when buying intent shows up.
Built-in CRM
Every conversation rolls up into a contact record: emails, phone, page history, tags, notes, deal stage. Export to CSV or push to your existing CRM via API.
Bring your own AI
Plug in your Gemini, OpenAI, Claude, OpenRouter or self-hosted key. You pay the LLM provider directly at their list price. No per-message AI markup, ever.
In-built RAG
Upload PDFs, paste URLs, drop in raw text. The platform chunks, embeds and indexes everything. The AI answers with citations and links back to the source.
MCP tools out of the box
Speak the Model Context Protocol as a client. Register any remote MCP server (Shopify, Stripe, HubSpot, Notion, Postgres, your own) and the AI can call its tools during conversations. Every call is audited, PII-redacted, and gated behind per-tool approval so a rogue upstream can't ship a stealth tool the agent will run.
How MCP support works →Bring your own LLM · we integrate with
GeminiGoogle Gemini
OpenAIOpenAI GPT
AnthropicAnthropic Claude
OpenRouter200+ models, one key
OllamaSelf-hosted, on-prem
Installs in minutes on WordPress (official plugin), Shopify (theme snippet), Webflow, Next.js, or any site with one script tag.
Your AI, your keys
We build the integrations. You own the model.
Most chat platforms force you onto their LLM and mark it up. OpenAgent connects to whatever you already pay for, or self-host. Switch providers in a click; no vendor lock-in, no per-message margin.
Gemini
Google AI Studio
OpenAI
GPT-4o, o-series
Anthropic
Claude 4.x
OpenRouter
200+ models, one key
Ollama
Self-hosted, on-prem
vLLM / TGI
Your GPU cluster
Don't see your provider?
Anything that speaks OpenAI-compatible chat completions works out of the box. Got something else? We'll wire it up.
Real-time alerts
Hot lead just landed. Your team already knows.
The moment a visitor sends a message, OpenAgent pings every channel you care about: Slack and Google Chat with the visitor name, the page they're on, and a one-click jump back to the conversation. Browser push, email, in-app and sound on by default.
Slack
Incoming webhook
Google Chat
Space webhook
Email
Per-user digest
Browser push
OS native
In-app
Live counter + chime
Sound
Per-tab chime
Per-agent quiet hours, on-call rotations and routing rules so the right person sees the right chat. Setup is a paste of the webhook URL. No OAuth dance, no bot install.
How it works
Live in 10 minutes. No SDR call required.
- 01
Drop the snippet on your site
One <script> tag. Bubble appears bottom-right. Visitors can chat in 60 seconds, even before you connect an LLM.
<script src="https://app.openagent.in/openagent.js" data-tenant="yoursite"></script> - 02
Plug in your LLM key
From the dashboard, paste your Gemini, OpenAI, Claude or OpenRouter key. Or point at your self-hosted endpoint. Test, save, done.
Provider: Gemini Model: gemini-2.5-pro API key: AIza•••••••••••••••••• [test] ✓ - 03
Train Aria on your docs
Drop in PDFs, paste a help-centre URL, point at a sitemap. RAG indexes everything. Answers come back with citations.
Indexed: 1,284 chunks across 42 docs Recall@5: 0.94 Avg cite confidence: 0.81 - 04
Let humans handle the hard ones
When the AI isn't sure, it escalates with full transcript, intent and a suggested reply. Agents pick up where Aria left off.
↑ Escalated to #support (low confidence) Reason: pricing override Aria's draft kept for review
Pricing
Honest pricing. No AI markup.
One flat platform fee per site. Pay your LLM provider directly for tokens. Cancel any time. Your data is yours, exportable with one click.
Save 40% with annual billing
Pro
$3/site /month
$36/site /year
billed annually
Bring your own LLM. Get the full platform.
One $36 payment per site per year. Prefer monthly? $5 per site per month, cancel any time. No card required to start.
- Unlimited live chat
- Unlimited agent seats
- Connect your Gemini, OpenAI, Claude, OpenRouter or self-hosted LLM
- AI agent with RAG (knowledge base, URLs, files)
- Bot to human and human to bot escalation
- Real-time visitor alerts (Slack, Google Chat, email)
- Built-in CRM and conversation export
- Smart routing
- Embeddable widget
- Up to 5 GB attachment storage per site
- 90-day conversation history
- Email support
Business
Custom
SLA, dedicated infra, white-label.
- Everything in Pro, plus:
- SSO (SAML / OIDC)
- Dedicated infra and SLA
- Audit log export
- Custom domain and branding
- Custom storage allocation
- Priority engineer Slack channel
Prices in USD. Pro is $5/site/month billed monthly, or $36/site/year billed annually. Your LLM tokens are billed directly by the provider you choose. OpenAgent never marks them up.
FAQ
Straight answers to the obvious questions.
Yes. $36 per site per year billed annually works out to $3 per site per month; on monthly billing it's $5 per site per month. Either way it covers the whole platform: unlimited live chat, AI agent, knowledge base, routing, analytics, every integration. The only other cost is whatever you pay your LLM provider for tokens, and that bill goes straight to them.
Out of the box: Google Gemini, OpenAI (GPT-4o and o-series), Anthropic Claude, OpenRouter, Ollama, and anything OpenAI-compatible (vLLM, TGI, LM Studio, Together, Groq). If your provider isn't on the list, we'll wire it up, usually the same day.
Keys are encrypted at rest with a per-tenant key, only decrypted in memory at request time, and never logged. You can rotate or revoke them from the dashboard. The platform falls back to live-chat-only the moment a key becomes invalid.
Tawk.to's live chat is free, but the AI Assist add-on bills you again once you cross the free response quota, plus paid hires and paid features on top. Intercom is great, but the AI alone runs around $0.99 per resolution and seats start at $74 a month. OpenAgent gives you live chat plus an AI agent on your own LLM keys from $3 a site a month billed annually ($5 monthly). Same surface, no margin stack on AI.
Yes. WordPress gets an official plugin: upload the ZIP from the install guide, activate, paste your workspace slug in Settings, done, no theme edits. Shopify gets a theme snippet that survives theme updates and links logged-in customers to their conversations automatically. Both are on the install guide at /install, along with notes for Webflow, Next.js and plain HTML.
No, OpenAgent is a hosted platform. Teams with data-residency or compliance constraints use the Business tier (dedicated infrastructure in the region you need, SLA, SSO, audit log export) and bring their own LLM keys, so model inference runs against the provider, endpoint, and region you choose, including a model you host yourself.
Multi-tenant by design: every row carries a tenantId derived from the authenticated session, never from the client. Data is stored on encrypted disks. You can request access, correction, export or deletion at any time by emailing [email protected].
Yes. One <script> tag, drop it anywhere. The widget renders in a Shadow DOM so it never collides with your site's CSS. It loads lazily after first paint and won't tank Lighthouse scores.
Yes. OpenAgent is a first-class MCP client. Register any remote Streamable-HTTP MCP server (Shopify, Stripe, HubSpot, Notion, Postgres, your own) and the AI can call its tools during conversations. Per-tool approval, PII-redacted audit log, and a rug-pull defense that catches an upstream server changing a tool description after approval. See /mcp for the feature page and /guides/how-mcp-works for the plain-English explainer.
Yes. Paste an incoming webhook URL (hooks.slack.com or chat.googleapis.com) into Dashboard → Alerts and every new visitor message lands in that channel with visitor name, page they're on, message preview, and a one-click link back to the conversation. Routing rules and per-agent quiet hours apply. Browser push, in-app pings, sound and email alerts are also on by default.
Pro is a flat per-site fee regardless of conversation volume. The only thing that scales with usage is your LLM bill, which you control. When you need SSO, audit log export or a dedicated cluster, Business kicks in.
Talk to us
Have a question?
We answer in person.
Drop your details and we'll reply within one business day, from a real human, not a sequencer. Or chat with Aria in the bottom-right; she's running on the same stack we sell you.
- Email[email protected]
- Phone+91 80754 95629
- Office91springboard, 45/3, Residency Road, Bengaluru 560025
- Live chatBottom-right bubble. Try it now.
Self-serve, 10 minutes
Provision a workspace from $3/mo
Paste a model key, drop in one doc, copy the snippet. No call required.
Or talk to a human first. Send a quick note below and Sujith will reply within a US business day.