Iris (AI)
Iris — your AI assistant
Iris is the AI assistant built into your Ngris dashboard. Describe the edge rule you want in plain language and Iris drafts it — a firewall rule, a rate limit, a full traffic policy, a routing rule, an error page, or a metric alert — then applies it under your own permissions after you approve. It also searches your HTTP traffic logs from a plain-English query, answers questions about your account (endpoints, members and roles, plan entitlements and quotas, verification) by reading the live API, and explains why an individual request was blocked or allowed.
Chatting with Iris requires a plan that includes the Iris AI assistant — see Pricing. On a plan without it, the chat shows an upgrade prompt instead.
Unlike a generic chatbot, Iris is wired into the real Ngris API. When you ask it to “turn on rate limiting for my API endpoint,” it drafts the rule, shows you a confirmation card, and only applies it after you approve. Every write it makes is recorded in an audit log you can review.
Opening Iris
There are two ways to chat with Iris, and they share the same conversation engine:
- The chat widget — the floating button in the bottom-right corner of every dashboard page. Click it to open a side panel without leaving the page you’re on. Iris knows which page you’re viewing, so “harden this endpoint” just works.
- The full chat page — a roomier, full-screen conversation at
/account/ai/chat, good for longer sessions and reviewing tables.
Some dashboard alerts include an “Ask Iris” link that opens the widget pre-loaded with the relevant context (for example, a rule you were just editing).
What Iris can do
Iris drafts edge rules from plain English, compiles traffic filters, and explains individual requests. A few examples:
| Area | Ask it to… |
|---|---|
| Firewall rules | “Block scanners hitting /admin” — Iris drafts the firewall rule with the right match and action for you to review. |
| Rate limits | “Limit login to 5 attempts per minute per IP” — drafts a scoped rate-limit rule with the right window and scope. |
| Traffic policy | Draft a WAF, JA3 bot-management, CORS, JWT-validation, redirect, header-rewrite, or caching rule. |
| Routing rules | Draft a routing rule that picks the right agent from your connected fleet for a match condition. |
| Error pages | Generate a full HTML/CSS error-page template with the error tokens filled in. |
| Metric alerts | “Alert me if p95 latency goes over 300 ms” — drafts a threshold alert for you to review and enable. |
| Smart-filter | Compile a plain-English description (“5xx on checkout slower than 800 ms, last hour”) into a real Traffic Inspector filter you can tweak and re-run. |
| Per-request explain | Open a captured request and ask why it was blocked or allowed — which rule matched, and a suggested fix. |
| Edit rules | Disable, delete, rename, or reorder existing rules — “disable the jwt-verify rule.” |
| How-to | Answer configuration questions from the docs and knowledge base, navigate you to the right page, and open a support ticket. |
If you ask for something Iris can’t do — account-inventory questions, a natural-language search of your stored logs, an outage narrative, or anything your plan or role doesn’t permit — it tells you plainly rather than guessing. It won’t invent endpoints, IDs, or API calls.
Confirmation & audit
Reads aren’t gated — but Iris never makes a change silently. Before any write — creating, updating, enabling, or deleting — it shows a confirmation card describing the action in your own terms:
- a plain-language title (e.g. “Enable rate limiting on api-prod”),
- a preview of exactly what changes — an update shows the current value → the new value field by field; a create shows the values that will be set; a delete shows which resource is being removed,
- a blast-radius preview for firewall, rate-limit, and traffic-policy rules — Iris replays your recent real traffic through the proposed rule and shows how many requests it would newly block, throttle, or redirect, how many of those were previously succeeding, the top affected IPs/paths/countries, and a safe / risky verdict, and
- Deny / Allow once / Always allow controls — nothing happens until you approve.
Every action Iris performs on your behalf is written to an audit log. Review the history of AI-initiated changes — what was done, when, and the outcome — any time on the AI activity page (Account → AI activity, at /account/ai-activity). This keeps the assistant accountable: there’s always a record you can audit.
| When | Action | Result |
|---|---|---|
| 10/19/2026, 2:02:11 PM | Create firewall / policy POST /api/firewall/policies |
Success |
| 10/18/2026, 9:47:30 AM | Update firewall / rate limit PUT /api/firewall/rate-limits/3f2a9c1e-5b7d-4e8a-9c21-0d6b4a8e7f13 |
Success |
| 10/17/2026, 4:20:05 PM | Delete firewall / policy DELETE /api/firewall/policies/8c41d2b7-0e9f-4a63-b5d8-2f7e1c9a4b60 |
Failed (403) |
Hands-off autonomy (opt-in, firewall-only)
By default Iris asks before every change. When you’re ready, you can let it handle one narrow, reversible job on its own: new firewall rules — blocking an attacking IP or range, or adding another firewall rule. This is the only kind of write Iris can make unattended; every other change (editing or deleting a rule, rate limits, traffic policy, routing) always comes back to you for approval. Autonomy is set on the AI Usage & Activity page under Iris settings (/account/ai-activity), per kind of change — for the firewall ones you choose:
- Ask me first — the default; a confirmation card every time.
- On probation — the first firewall change applies on its own and must verify healthy before further ones auto-apply; while one is being checked Iris asks first. Once one passes, further changes apply on their own for the probation window you set (60 minutes by default).
- Act on its own — applies unattended, but only when the traffic dry-run says it’s safe.
Two independent opt-ins are always required: your plan must include the autonomy entitlement and an account owner must switch it on — and even then, anything risky (or anything Iris can’t fully simulate) still comes to you. Every automatic firewall change is verified against real traffic and reverted on regression; too many reverts in an hour and it switches itself back to ask-first. Autonomy settings are human-only — Iris can never change its own permissions. Changing them needs a role that can write the account's settings.
Proactive suggestions
Iris also flags things that need attention on its own — like a certificate, domain or card about to expire — as Iris suggestions on your dashboard, and opens the chat for critical ones. To mute them for everyone on the account, switch off Proactive suggestions under AI Usage & Activity → Iris settings; switch it back on any time.
Iris settings
How much Iris does without being asked, for everyone on this account. Only people can change these — Iris can't change its own settings.
| Change | Iris | Probation window |
|---|---|---|
Block an attacking IP or range A new firewall rule that blocks traffic Iris identified as an attack. |
min |
|
Other new firewall rules New allow or other firewall rules. Editing an existing rule always asks first. |
— | |
Remove a rule that blocks good traffic Deleting a firewall rule. |
Always asks first | — |
Rate-limit changes Creating or tuning rate-limit rules. |
Always asks first | — |
Traffic-policy changes Creating or editing traffic-policy rules. |
Always asks first | — |
On probation: Iris applies one change of this kind on its own and waits for it to pass its traffic check; while that check runs it asks first. Once one passes, further changes apply on their own during the probation window.
Metric alerts from plain English
Describe a threshold you want to watch and Iris drafts a metric-alert rule for you to review and enable — for example “alert me if p95 latency goes over 300 ms”, “warn me when my error rate crosses 1%”, or “tell me when I hit 80% of my data cap”. Iris fills in the metric, the comparison and threshold, the evaluation window, and the endpoint or account scope; you adjust and turn it on.
A metric alert is a notification, not an autopilot: when the condition is met it fires so you know to look. Iris does not silently reconfigure your account to “fix” a latency breach or an error spike, and there is no continuous objective it reconciles on its own. It drafts the alert; you decide what to do when it fires (and it can help you draft the firewall or rate-limit rule you reach for, each behind its own confirmation card).
Permissions, privacy & limits
- Runs as you. Iris acts under your authenticated session and role. It can never exceed your own permissions, and team members only see and change what their role allows.
- Hard server-side limits. Beyond your role, the platform enforces what the assistant may attempt at all: it can’t touch admin areas, and irreversible account-level actions (deleting the account, transferring ownership) are deliberately left to you to do by hand — these are blocked on the server, not just discouraged in the prompt.
- Confirmation before writes. Reads are free; every change is gated behind an explicit confirmation card with a preview (see above).
- Rate-bounded. The number of changes Iris can make per hour is capped per plan, so an unexpected loop can never run away.
- Usage limits. Your plan sets an hourly Iris message limit and a monthly AI token budget shared by your account’s members (usage and what is left:
/account/ai-usage). Each step of an answer is checked against what is left before it is sent: a message whose first step wouldn’t fit is refused up front, and a long multi-step answer stops when the budget runs out part-way instead of running past it — both answer “Monthly AI token budget reached” (HTTP 429). The budget resets on the 1st of the month. Iris sends its instructions and the earlier conversation again on every step; the part served from the AI provider’s prompt cache counts at 10% toward the budget, the rest in full. A refusal or error shows in the chat under your message and is still there when you reload the page or reopen the chat; it is not sent to Iris as part of the conversation. - New chats per hour. Starting a new chat (
POST /v1/ai/sessions) is limited to 30 per user and 120 per account per hour; continuing an existing chat doesn’t count. Past either cap the request answers HTTP 429 with aRetry-Afterheader and{"code": "ai_session_rate_limited", "scope": "user" | "account", "limit", "retry_after_seconds"}. - Audited. AI-initiated changes are recorded on your account’s AI activity page.
- Plan-aware. Iris respects your plan’s entitlements and limits — if a feature isn’t available on your plan, it says so instead of attempting it.
For the full marketing overview of the assistant, see the AI features page.