AI
Knowledge base RAG
Turn your docs into accurate answers.
- Upload PDFs, Markdown, CSV, JSON, text
- Hybrid semantic + keyword search with reranking
- Confidence score on every single reply
AI that knows when to step aside AI when it can.
AI when it can.
Humans when it matters.
Zybo AI pairs a RAG-powered AI agent trained on your docs with a full human agent inbox. Escalation, routing, priority, internal notes, and real-time handoff — one platform, no stitching.
< 100ms First-token streaming latency
3 modes AI · AI + Escalation · Human
Every reply Confidence-scored before it ships
✦
✧
✦
LIVELOW
Hi! Ask me anything — I can answer from your docs.
Confidence 0.92
High match — 3 sources cited
Agent Stats
Alice 3 / 5
Marcus 5 / 5
Priya 1 / 4
Platform
Everything a modern support stack needs —
in one place.
The AI, the inbox, the queue, the knowledge base, and the ops tooling teams normally bolt together from six vendors — built as one system, with workspace isolation baked into the schema.
Customer asks
AI answers from your docs
Every reply is scored
Human on standby
Resolved
COMMUNICATION
Real-time chat widget
Go live with a single script tag in your website.
- Streaming replies and typing indicators for a modern feel
- Image attachments via direct upload
- Returning-visitor recognition out of the box
Human agent inbox
Full control over every conversation.
- Claim, reassign between teammates, release, or hand back to AI
- Queue · mine · supervisor views
- Live updates over Socket.IO — never poll
Smart escalation
AI hands off the moment it should.
- Triggers: user ask, low confidence, supervisor
- Auto-routes to the least-busy available agent
- Reason + score saved on the timeline
Auto-resolve idle chats
Close the loop without lifting a finger.
- AI checks in, then resolves quiet or "thanks, bye" chats
- Reopens instantly the moment the visitor replies
- Per-AI-agent nudge + close timers you control
Priority & tags
Never lose the chat that matters.
- Four priority levels, org-scoped tags
- Filter the queue by tag in one click
- 60-second sweep auto-bumps stale chats
Notifications that travel
Never miss the chat that needs you.
- In-app center: assignments, escalations, @mentions, billing, KB
- Mobile push the second a chat needs a human
- Per-category controls + sound, all in one place
AI copilot for agents
Reply faster with AI in the seat next to you.
- One-click reply suggestions
- Thread summaries on arrival
- Intent detection per message
Internal notes + @mentions
Collaborate without the customer seeing.
- Team-only notes alongside the transcript
- @mention teammates inline
- Hidden from the AI context and the user
End-user context
Know who you are talking to, instantly.
- Identity, geo, device, browser, UTM source
- Prior chat history + identify() SDK for logged-in users
- Geo resolves locally in under a millisecond
Website ingestion
Point at your site — we turn it into KB content.
- Public URL crawl with include/exclude patterns
- Re-ingest a single page anytime to refresh it
- Up to 30 pages on Starter, 100 on Pro per run
Visitor map
See where every visitor is talking to you from.
- World map + top-10 cities, sized by unique visitors
- Engaged-visitor layer on Starter, all-visitor layer on Pro
- Privacy-first: geo + device class only, respects Do-Not-Track
Slash templates
Canned replies that never sound canned.
- "/" command palette in the composer
- Variables like {{user_name}}, {{agent_name}}
- Resolved per conversation
Append-only audit log
Every change, traceable forever.
- Every mutation + auth event logged
- Actor, IP, and metadata captured
- Queryable per workspace
Agent actions
Your AI agent can act, not just answer.
- Call your own APIs — no code, any vertical
- Verified visitor identity for "my orders"
- Approval gates + full audit on writes
AI Agent Modes
Pick a mode. Match the moment.
Every AI agent picks its own mode. Run a self-serve FAQ agent, a hybrid support agent, and a human-only sales agent side by side in the same organization.
01
AI Only
The AI handles everything.
Answers come straight from your knowledge base, with citations. The AI politely declines what it does not know and never escalates. Best for FAQs, docs, and onboarding.
- Instant answers · no wait time
- Confidence score on every reply
- Runs 24/7 with zero agent staffing
02
AI + Escalation
AI-first, human safety net.
AI replies by default, then hands off on three triggers — user request, confidence below your threshold, or a supervisor stepping in. The conversation auto-routes to the least-busy available agent.
- Three escalation triggers wired in
- Auto-assign to the least-busy agent
- Reassign back to AI mid-conversation
03
Human Only
Every chat goes to a person.
No AI processing — messages flow straight into the queue. Zero hallucination risk for billing disputes, complaints, or live sales — and your agents still get priority, tags, and copilot.
- Zero hallucination risk
- Full human control from hello
- Priority, tags & copilot still included
How it works
Live in 4 steps.
Sign up to first real chat in under ten minutes. No engineers, no six-month rollout, no SDK to compile.
01
Create your account & AI agent
Verify your email, name your AI agent, pick a mode. Sixty seconds in, your workspace is ready.
02
Upload your knowledge base
Drag in PDFs, Markdown, CSV, JSON, plain text — or point at a public URL and we crawl it for you. Zybo chunks, embeds, and indexes everything automatically.
03
Get your install key
One click in AI Agent → Integrate. Copy your key — it locks the widget to this single AI agent and is shown only once.
04
Drop the widget on your site
Paste one script tag in your site footer. Customers can chat instantly; your team can take over any time.
What your customers see
No app to install, no account to create. The widget appears in the corner of your site — they click, they ask, they get an answer.
Agent Inbox
The inbox your team actually wants to open.
Three views — priority queue, my conversations, supervisor oversight. Real-time everywhere, with the context panel, timeline, and copilot one keypress away.
- Queue auto-sorted by priority and wait time
- Identity, geo, device, and history in the context panel
- Timeline of every event from start to resolve
- Workload stats: capacity, response time, availability
- Internal notes, @mentions, and a notification bell
Mobile app
Your agent inbox, in your pocket.
Zybo AI travels with you. Watch the AI handle chats live, get a push the moment a conversation needs a human, and take over, reassign, or resolve — without opening your laptop.
- Live inbox over the same realtime channel as the web — new chats and replies stream in instantly
- Take over, escalate, reassign, resolve, or hand back to AI from your phone
- Push notifications the second a chat escalates or is assigned to you
- Magic-link & Google sign-in, with multi-workspace switching
- AI conversations stay live-viewable — step in only when it matters
Coming soon
One inbox for every channel.
The same AI answering and human handoff you run on your website — soon across the channels your customers already message you on.
- Messenger
0 AI agent operation modes
0 escalation triggers
0 priority levels
0 script tag to install
Loved by support teams
Why teams pick Zybo AI.
A few words from the people building support stacks they actually want to live in. (Placeholders — replaced with real quotes before launch.)
FAQ
Questions, answered.
Everything you need to know before starting. Still stuck? Talk to sales.
How long does setup actually take?
Under ten minutes for most teams. Verify your email, name your AI agent, drop in a few PDFs or point at your website URL, copy your install key, paste one script tag on your site. No engineers, no compile step, no SDK to learn.
What happens when the AI doesn't know the answer?
It doesn't guess. Every reply gets a confidence score. If the score drops below your threshold (0.3 by default — you can tune it), the chat hands off to a human agent automatically, with the score and the reason saved on the timeline.
Will my data ever mix with another company's?
Never. Every workspace is isolated at the database level — different conversations, different documents, different AI agents, all walled off. Cross-tenant leakage is mathematically impossible, not a policy.
Can our support team take over a chat from the AI?
Yes — mid-conversation, with the full transcript, customer history, and context already in front of them. They can also reassign to another teammate, release the chat back to the queue, or hand it back to the AI when it's a question the agent can handle.
What can I upload as the knowledge base?
PDFs, Markdown, CSV, JSON, and plain text — or point at a public URL on Starter+ and Zybo crawls your website for you (with optional include/exclude patterns). Each format is chunked intelligently (sections, paragraphs, sentences, rows) and embedded for hybrid semantic + keyword search.
How fast do customers see the AI reply?
First token under 100 ms — replies stream in word by word, like ChatGPT, instead of making the customer wait for a full response. Typing indicators are real, not faked.
Can customers send screenshots or attachments?
Yes — images (JPEG, PNG, WebP, GIF). Up to 2 MB per file on Starter, 10 MB on Pro. Free plan keeps things text-only. Files upload directly to secure storage; we never see the bytes.
What if my agent's wifi cuts out mid-chat?
A 60-second grace timer kicks in. Reconnect inside the window and their chats stay assigned. Miss it and the chats auto-release back to the queue, the agent flips to offline, and the next available teammate picks them up.
Give your support team an unfair advantage.
Ship an AI-powered inbox your customers actually like — and your agents never want to leave. Free to start, ten minutes from sign-up to first real chat.
No credit card · free tier forever · cancel anytime