Customer support, finally autonomous

Support that resolves.

Luna is the AI customer-support agent that handles email, Slack, chat and API end-to-end — in any language. Ships with memory, full observability, and human-in-the-loop control. Whitelabel it as your own.

Invite-only beta · onboarding our first design partners
app.lunahq.io · inbox
Resolving
M
Inbox
Your tasks2
Mentions4
All14
Unassigned0
Channels
Slack5
Email6
Web chat2
API1
Views
Auth issues
Billing
Feature requests

All conversations

14 active · last 24h
Login issue — SSO keeps looping
Our whole team is locked out, SSO keeps looping…
Resolving 1m
Card declined on annual renewal
Tried to renew but my card keeps getting declined…
Researching 4m
Billing inquiry — possible double charge
I was charged twice this month for my Pro plan…
QA 12m
‫استفسار عن خاصية التصدير‬
‫أبغى أعرف وش الفرق بين الخطة الاحترافية…‬
Resolving 18m
Account locked — lost 2FA device
Cannot access account after losing my authenticator…
Resolving 32m
Feature request: dark mode for portal
Several customers have been asking for dark mode…
Replied 1h
Cancel subscription mid-cycle
Need to cancel before the next billing date but want…
Resolved 2h

SSO loop — TechHub locked out

SUP-0084 · Slack · URGENT
Intake 58s

Classified · AUTH · URGENT · 5 min ETA

Research 45s

SSO certificate expired May 11. ACS URL mismatch after renewal is the likely root cause.

Reply sent 30s

Luna replied with three checks + workaround.

Resolve just now
Rotating cert
What Luna does

An agent that finishes the work.

Not a chatbot that pings a human at the first sign of friction. Luna handles the full lifecycle of a customer conversation — research, resolve, follow up, close — and escalates only when it should.

01 · Resolves

Handles repetitive queries end-to-end.

Auth issues, billing questions, refunds, account changes, feature requests — Luna researches, drafts, gets the action done, and follows up. Works in any language the customer wrote in.

Amelia
Our whole team is locked out, SSO keeps looping 😩
Luna
Your SSO cert expired May 11. I've rotated it and applied the fix — you should be able to sign in now. Verifying with your account.
02 · Escalates

Pauses when it's not sure.

Luna writes back when it's confident, pauses when it isn't. Refunds over a threshold, security-sensitive actions, ambiguous tone — it asks a human, with full context, instead of guessing.

// confidence threshold if action.risk > 0.7: pause(reason="refund > $500") notify(@mohanad, context=full) # waiting · 0:32s ✓ approved · resuming
03 · Learns

Gets sharper every cycle.

Each resolution writes back to Luna's reflective memory — episodic facts, semantic policies, customer preferences. Yesterday's edge case is today's instinct.

// memory layer episodic TechHub uses Okta SSO ×14 semantic renew cert → restart IdP SOP policy escalate auth → senior ±0 tone Amelia prefers concise ×8
Inside Luna

Built like a brain.
Not a chatbot.

A chatbot answers one message at a time and forgets you the moment it's done. Luna is a whole reasoning system — it perceives, remembers, decides, acts in your tools, and gets sharper every time. Here's the difference.

A normal AI chatbot
message 1 model · 1 prompt reply
  • Forgets every conversation
  • Can't touch your tools or data
  • Guesses confidently when unsure
  • One channel, breaks on the 2nd turn
vs
Luna
perceive recall reason act reflect
  • Remembers every customer & decision
  • Acts in 26 tools — Zendesk, Slack, Stripe…
  • Pauses for a human instead of guessing
  • Every channel, any language, fully logged
One core. Six faculties. Wired together, always thinking.
Luna
Reasoning core
thinking
Perception how it hears you omnichannel
voice· chat· web· email· API· any language
Memory what it remembers always on
per-customer· episodic· semantic· reflection
Reasoning & judgment how it decides deliberate
plans steps· weighs confidence· pauses, won't guess
Skills & tools what it can do 26 + MCP
Zendesk· Slack· Stripe· Linear· +22 native· any MCP
Sub-agents how it scales big jobs on demand
spawn delegate merge
Reflection how it improves self-sharpening
reviews every resolution· writes back to memory
Every single request runs the full loop — never a one-shot reply
00
Perceive
hears the request, any channel
01
Recall
pulls memory + context
02
Reason
plans · checks confidence
03
Act
runs tools · sub-agents
04
Reflect
learns · writes to memory
Woven through everything — the security you don't have to build
Permissions & roles
scoped · least-privilege
Secrets vault
AES-256 · per-tenant
Private memory
encrypted · isolated
Identity & SSO
SAML · OIDC · SCIM
Audit trail
every action logged
Reasoning core Faculty Signal in Action out
Trust, built in

The parts that are hard to get right.

The result is an agent you can run in production — not a demo that falls apart on the second turn. Every Luna deployment ships with the same guardrails on day one.

Event-level observability.

Every tool call, every prompt, every retry — sealed in an append-only log. Replay any conversation. Audit any decision. Nothing happens silently.

12:04:01INTAKESUP-0084 · classified AUTH · URGENT
12:04:03RESEARCHqueried knowledge · 3 sources matched
12:04:08TOOL_OKslack.lookup_thread(channel=#support)
12:04:12RETRYokta.fetch_audit · 429 · backoff 2s
12:04:14TOOL_OKokta.fetch_audit · 47 events
12:04:20REPLYposted to Amelia · 3 recommendations
12:04:23HITLresolve step · awaiting approval
12:04:31APPROVEDby mohanad · cycle resumed
12:04:39RESOLVEDSUP-0084 · 38s · cert rotated

Encrypted secrets at rest.

API keys, tokens and connection strings live in a per-deployment vault. AES-256-GCM at rest. Per-tenant scope. Never logged.

// .env never touches disk vault.set("OKTA_API_KEY", sk) // → scope: techhub · prod // → sealed: aes-256-gcm

Reflection-driven memory.

Each conversation writes back what was learned. Luna accumulates operating knowledge, not just transcripts.

episodic"TechHub uses Okta SSO"×14
semantic"renew cert → restart IdP"SOP
policy"escalate auth → senior"±0

Human-controlled recovery.

When Luna isn't sure, it pauses and asks. No autonomous guesswork on production work.

PAUSElow confidence: refund > $500
WAIT@mohanad approval0:32s
RESUMEapproved · resuming cycle

Zero silent retries.

Failures surface immediately, with the full context that caused them. You see every retry, every backoff, every fallback path.

// every retry is a record attempt 1 · 429 · backoff 2s attempt 2 · 429 · backoff 4s attempt 3 · 200 · 1.2s // → surfaced in event log

Per-tenant isolation.

Every tenant's data, memory and keys are walled off. A privacy classifier gates anything shared — and defaults to tenant-only, always.

// classifier · before any shared write "TechHub uses Okta SSO"tenant-only "lead with policy, then a path"global-safe // on any doubt → tenant-only

Fails closed, not open.

After repeated failures Luna locks the agent — no token burn, no degraded autonomous guessing. It waits for a human, not a retry storm.

// 3 consecutive failures agent.stateLOCKED // refusing calls · 0 tokens spent // → operator unlock required
Whitelabel

Same engine. Your brand.

Drop in your logo, fonts and colours. Wire it to your Slack, your inbox, your help centre. Your customers see your product — not ours. Luna runs underneath, invisibly.

T TechHub Support
Online
AP
Our team is locked out — SSO keeps looping.
T
Hi Amelia — checking your SSO config now. One sec.
Embedded in TechHub.io Powered by Luna
a Auroracare
Concierge
JC
Trying to reschedule my consultation — site won't let me past step 2.
a
Hi Jamie — I've cleared the form for you and reserved Thursday 3pm. Please confirm.
Embedded in aurora.care Powered by Luna
$_ flowops/help
● Live
MR
Pipeline failing on stage-3, getting 502 from webhook target.
$
Webhook timing out — your stage-3 target's TLS cert expired 11 min ago. Rotating now.
Embedded in flowops.dev Powered by Luna
How it works

From idea to deployment in three moves.

Configure once. Wire your channels and tools. Ship it from the hub. The framework handles the plumbing so your team can focus on the part that actually matters — the behaviour, the brand, the resolution rate.

Configure

Brand, tone, channels, tools, knowledge sources. Inherit secrets and observability from the core.

// agents/support.luna.ts agent({ name: "Luna · TechHub", channels: [slack, email, web], tools: [okta, stripe, notion], })

Deploy

Push to the hub. Luna provisions an isolated runtime — separate vault, memory store, event log — under your branded subdomain.

$ luna deploy techhub ✓ vault sealed ✓ memory provisioned ✓ event log armed → help.techhub.io · live

Iterate

Watch every conversation in the hub. Inspect every decision. Edit memory inline. Pause when you need to.

// inspect & tune support.pause("checking a reply") memory.add("escalate auth > senior") ✓ applied to next cycle

Most AI tools are wrappers around a single model with a single prompt. Luna ships agents you can actually trust in productionnot demos that fall apart on the second turn.

The LunaHQ manifesto
Pricing

Pay for resolutions, not seats.

Every plan includes the full Luna core: vault, memory, observability, HITL, whitelabel. You only pay for conversations actually handled.

Starter

$49 / mo

For small teams testing Luna on a single channel.

  • 1 deployed agent
  • 500 resolved conversations / mo
  • 14-day event retention
  • 2 channels (Slack + email)
  • Email support
Request access

Enterprise

Custom

Private deployment, custom retention, dedicated support and compliance alignment.

  • Unlimited conversations
  • Private deployment available
  • Direct line to the founder
  • SOC2-ready · HIPAA-aligned · DPA available
  • Custom integrations
Talk to us

Ship support that resolves.
Not another chatbot.

Luna is in invite-only beta. We're working with a small group of teams running production AI support — apply for access and we'll get back within a day.

Invite-only beta · no credit card required