ProductFeaturesAgentsSecurityPricingIntegrationsCompareBook a demo
Product overview

One graph. A calm briefing. A short, trusted queue.

Aanty is a typed-object conversation graph with an AI layer that processes what it's permitted to read, so you read only what needs you. Here's how the pieces fit together.

Default home

The Briefing replaces the channel list — and Needs You is the trusted queue beneath it.

Open aanty and you land on a per-person, AI-composed summary: what changed, what needs you, what was decided, what's blocked, what agents did, and what can safely be ignored. Below it sits the Needs-You queue — a short, deterministic list where every item carries a reason and an inline action, so you can approve, reply, or snooze without opening the channel.

  • Never a channel list. The default view is composed, not browsed — read-only, drill-down to the source thread in one tap.
  • Why-you is always present. A deterministic reason — named approver, directly asked, assigned to you — precedes every item; AI can promote or demote, never suppress a policy-driven delivery.
  • The badge counts what needs you, never what's unread.
Needs You
Why you're seeing this
Approval#aurora › risks
Approve risk-mitigation budget
kkumty-bot · why you: you're the named approver
Question#portal › auth
Which auth provider should the client portal use?
SSara Okafor · why you: directly asked you
Every item carries a reason. No reason, no interruption.
Explore the Briefing
The substrate

Every message is one of eleven typed objects, not a blob of text.

Type is assigned at write time (AI-suggested, author-overridable) and drives routing, notification priority, search facets, and analytics. Nobody else can retrofit this — it has to be the unit of communication from day one.

MessageQuestionAnswerDecisionUpdateTaskApprovalAlertDocEventCelebration
Explore typed messages
Entity-linked threads

Any object in any of your tools can own a conversation.

An invoice, a task, a risk, a course, an application, a document — any external object can be the anchor of a thread. The thread renders a live entity card (status, owner, actions) right where you're talking, and the source platform shows the discussion back. Events from Kafka topics, Redis streams, and webhooks arrive as system messages you can reply to, turning every state change into a place to talk.

  • One generic envelope. linked_entity {platform, entity_type, entity_id} — every sibling's objects map onto it.
  • Live, not a snapshot. Cards fetch through the source platform under your permissions; a stale card is flagged, never silently wrong.
  • Denied cards stay locked. If you can't see the source object, the card renders a locked empty state — never a partial leak.
kumty · risk High

Vendor SLA breach

Rendered inline from kumty's own data — status, owner, and actions stay live because the card fetches through the source platform, never a stale copy.

  • Actions run in the source — Mitigate or Escalate call back into kumty, never a local copy of the record.
Search, trimmed by permission at query time

One index for messages, files, canvases, and transcripts.

Keyword operators (from/in/has/before/after), semantic search, and facets by type, entity, and person — sub-second at scale. Permission trimming happens at query time, not after: a query can't surface what you can't see.

Keyword operators

from:, in:, has:, before:, after: — precise search when you know what you're looking for.

Semantic search

Find the right thread even when you don't remember the exact words.

Faceted by type & entity

Narrow by message type, linked entity, or person without writing a query.

Trimmed at query time

Results are scoped to your access before they're ranked — never filtered after the fact.

See search in depth
Agents are principals

Humans and AI agents are both first-class — with budgets and receipts.

Agents have identity, capability scopes, per-tenant budgets, and an audit trail, and every agent message shows its model, confidence, and citations. A human is never left guessing who — or what — they're talking to.

Unmistakably an agent

A persistent badge and its own accent colour. Never styled to impersonate a human.

Scoped & budgeted

Read-only → draft → low-risk write → consequential write with mandatory human approval.

Provenance on every message

Model, version, confidence, and cited sources — one tap opens the receipts.

MCP-native, both ways

Your graph is an MCP server under your ACLs; agents in aanty can use any MCP server.

How agents work
Book a demo

See what your team stops reading.

Bring a noisy channel. In fifteen minutes we'll show you the same day as a calm briefing and a short queue of what actually needed you.