AI assistance
AI that's already in your plan.
Answers your customers directly and helps your agents move faster. Built on Anthropic Claude and Voyage, included in every tier, off whenever you want. No per-resolution fees, no add-on line items, no surprise bills.
Customer-facing AI
Answer customers before a ticket is ever created.
The same knowledge base that grounds your agents' drafts can answer your customers directly. A chat-widget bot, an "Ask AI" panel in the customer portal, and live article suggestions as someone types a request all resolve common questions instantly, grounded in your articles and cited, with a one-tap handoff to a human whenever it's needed.
Every question answered this way is a ticket your team never has to open. Deflection runs on the same Anthropic Claude and Voyage stack as agent-assist, so the answers stay accurate, on-brand, and grounded in your content, not invented.
- Chat-widget bot answers from your knowledge base, grounded and cited
- Portal "Ask AI" lets customers self-serve before filing a ticket
- Pre-submit suggestions surface relevant articles as a request is typed
- One-tap escalate-to-human, with full context, never a dead end
- Same RAG engine (Claude + Voyage) as the agent-side AI
Ask AI · customer portal
"How do I reset my API key?" The bot answers from your "Managing API keys" article, links the source, and offers to connect a human if the customer still needs one.
Three places customers get answered
Turn each one on independently. All three draw from the same knowledge base and hand off to a human the moment self-service isn't enough.
Chat-widget bot
The same one-line website widget that runs live chat can answer first with AI. Customers get an instant, cited answer from your knowledge base; if they're not satisfied, the conversation escalates to a live agent or a ticket without losing context.
Portal "Ask AI"
An "Ask AI" panel in the customer portal lets signed-in customers ask in plain language and get a grounded answer drawn from the knowledge they're allowed to see, before they ever raise a request.
Pre-submit suggestions
As a customer types a new request, Sysflows surfaces the articles most likely to answer it. Many close the tab with their answer; the rest submit a request that's already better described.
What's included on every plan.
These features are on by default when you sign up. You can turn any of them off from Settings at any time.
Draft replies
AI reads the conversation and your knowledge base, then writes a draft reply, with a citation chip showing which article it drew from. You accept, edit, or discard.
Auto-triage
Incoming tickets are scanned for team routing and priority. The AI proposes both. Your rules override, and agents confirm before anything is applied.
Thread summaries
Long conversations get a plain-language summary so a new agent can pick up without reading the whole thread. One click, always human-readable.
Resolution summaries
When a ticket closes, the AI writes a short summary of what happened and how it was resolved. Useful for knowledge base drafts and team reporting.
Similar resolved tickets
Before an agent replies, Sysflows surfaces tickets that were resolved the same way before. Context without searching, using semantic matching, not keyword lookup.
Semantic knowledge search
Your knowledge base is indexed with Voyage embeddings. Search finds the right article even when the words don't match exactly. Results appear alongside the composer.
Spam classification
Junk, marketing, and bulk mail is flagged automatically with a one-click banner to close silently. Repeat offenders are suppressed. Your queue stays clean.
AI knowledge drafts
Turn a resolved ticket into a knowledge article draft with one click. The AI writes the first version from the conversation. You review and publish.
RAG-grounded drafts
Drafts that cite your own articles, not the internet.
The AI reads your knowledge base before it writes anything. Every draft reply comes with a citation chip showing which article it drew from. If the source is wrong, agents catch it before it sends.
That's how RAG (retrieval-augmented generation) works in practice: the AI is grounded in what you've written, not in general internet knowledge. Answers stay accurate and on-brand.
- Grounded in your articles, not the open web
- Source cited on every claim
- Agent accepts, edits, or discards before anything sends
- Token usage shown inline, updated in real time
AI draft · grounded reply
A drafted reply confirms the customer is within the 30-day window and the refund has started, grounded in your own article, cited so an agent can verify the source before sending.
AI Agents, available on Growth and Pro.
AI Agents are named workers you configure. They can triage, draft, and act, but only within the limits you set, and only on the skills you trust them with.
Named agents with defined roles
Give an AI agent a name, a role (ticket assistant, chat assistant), and a set of skills. Each agent is distinct. You can have one for each team or workflow if you want.
Per-skill trust: auto-apply or propose
For each skill (triage, draft reply, link article, close ticket) you decide whether the agent acts immediately (trusted) or queues its suggestion for a human to approve (propose). You're in control of every capability independently.
An approval surface for every action
All AI Agent actions go through an "AI tab" on the ticket where agents can review, accept, or reject what the AI proposed or did. Nothing is hidden. The full action log is there.
Run-ceiling protection
Each AI Agent has a loop ceiling: a maximum number of autonomous steps per run. This prevents runaway automation. If the ceiling is hit, the agent stops and flags for human review.
Always optional
AI Agents are opt-in. You can run Sysflows entirely with the assistive AI features (drafts, summaries, triage proposals) and never configure an autonomous Agent. It's your call.
You stay in control
Every AI feature has its own switch. There's a master switch too.
Not comfortable with AI drafts? Turn them off. Want summaries but not auto-triage? That's one toggle. Want to switch AI off entirely while you evaluate? One switch at the top of the AI settings page.
Nothing the AI proposes is applied without a human confirming it, unless you explicitly trust a specific AI Agent skill to act autonomously. The defaults are conservative.
- Master on/off switch for the entire AI system
- Per-feature toggles: drafts, triage, summaries, search, spam, each independent
- Every suggestion human-confirmable before it applies
- AI Agent skills individually set to "propose" or "auto-apply"
- Token usage tracked and displayed in real time
- 80% and 100% token alerts, so you know before you run out
AI settings
Token allowances, in plain English.
Every plan includes a fixed token allowance per seat per month. Tokens are pooled across your whole team, so a quiet seat shares its budget with a busy one.
Starter · ~30 AI resolutions
Growth · ~80 AI resolutions
Pro · ~165 AI resolutions
What's a token? Tokens are the units AI models use to measure input and output text. One AI resolution, a complete draft reply exchange, uses roughly 10,000 tokens. The numbers above are estimates; actual usage varies with conversation length and which features you use.
Overage is opt-in only. When your team hits 80% of the allowance, Sysflows sends an alert. At 100%, AI features pause unless you've enabled a top-up. Top-ups cost $20 per 1M tokens and are never charged automatically. You decide before you spend.
Pooled across your team
The allowance isn't per-agent. It's per seat per month, pooled. A 5-seat Growth plan has 3.75M tokens in total. High-volume agents can draw more; quiet agents contribute their share.
80% and 100% alerts
You get a notification at 80% of the monthly allowance and again at 100%. No surprise bills. AI assistance pauses at the cap. It never auto-charges for overage unless you've opted in.
Built on Anthropic Claude and Voyage.
The technology behind the AI features is not proprietary black-box tooling. We use well-regarded, well-documented models you can read about yourself.
Anthropic Claude
Generative tasks (draft replies, summaries, triage proposals, knowledge drafts) are handled by Claude (Haiku for fast tasks, Sonnet for more complex ones). The model selection is automatic and cost-aware.
Voyage embeddings on pgvector
Semantic search, finding the right knowledge article or similar past ticket, uses Voyage embeddings stored in pgvector. Search finds meaning, not just matching words.
Common questions about AI
Is the AI on by default, or do I have to enable it?
AI assistance is on by default when you start a trial. You can turn individual features off, or switch everything off with the master toggle, from the AI settings page at any time. No features auto-apply anything without an agent confirming first.
Do the AI draft replies actually sound like us?
Drafts are grounded in your own knowledge base articles and the conversation thread, so they reflect your content and tone. The more you build out your knowledge base, the more on-brand the drafts become. Agents always review and edit before sending.
What happens when we hit the token cap?
At 80% usage you get an alert. At 100%, AI features pause. The platform keeps working normally, agents just won't see AI-generated drafts or summaries until the next billing period or until you enable a top-up. No surprise charge, no auto-overage.
Is our conversation data used to train models?
No. Your data is in your own isolated database and is not used to train any AI model. AI calls are made to Anthropic's API and Voyage's API under terms that do not permit training on customer data.
Can customers use the AI themselves, or is it only for agents?
Both. Alongside the agent-assist features, Sysflows includes customer-facing AI: a chat-widget bot, an "Ask AI" panel in the portal, and article suggestions shown as a request is being typed. Each answers from your knowledge base, cites its source, and escalates to a human when self-service isn't enough. Every question it resolves is a ticket your team never opens. You can enable any of these surfaces independently or leave them off.
What's the difference between AI assistance and AI Agents?
AI assistance (drafts, summaries, triage, search, spam) is available on every plan and always proposes. An agent confirms before anything happens. AI Agents are configurable autonomous workers on Growth and Pro that can take actions automatically, within the limits you set per skill. Both are optional.
How much does a top-up cost?
Opt-in top-ups cost $20 per 1M tokens, roughly $0.18 per AI resolution. They are never charged automatically. You enable top-ups in Settings, and you can set a hard cap on how many top-ups can be purchased per billing period.
Start free. AI included.
No credit card. 14-day full trial. AI on from day one.