What is next-best-action decisioning?

Next-best-action (NBA) decisioning is a real-time AI approach that selects the single best offer, message, or treatment to show each customer on each channel, based on their data, context, and predicted response. KaireonAI combines eligibility rules, ML scoring, contact policies, and multi-objective arbitration to return a decision in a single real-time API call.

Is KaireonAI really open source?

Yes. KaireonAI is Apache 2.0 licensed and self-hostable on any Linux, Docker, or Kubernetes environment. The full platform — visual decision flows, ML scoring, connectors, dashboards, and 160+ MCP tools — is free to use without limits.

What makes KaireonAI different?

KaireonAI ships a full next-best-action platform as open source: visual decision flows, 9 ML scoring algorithms, 50+ data connectors, contact policies, multi-objective arbitration, A/B experiments, and real-time dashboards. It is AI-native from day one with 160+ MCP tools, so you can build and maintain decisioning config in natural language from Claude Code or Cursor. Self-host anywhere with Docker or Kubernetes.

What channels does KaireonAI support?

KaireonAI is channel-agnostic. Any channel that can make an HTTP request — web, mobile, email, SMS, push, in-app, contact center, ATM, kiosk — can call the Recommend API and receive a ranked list of offers with explainable decision traces.

How fast is decisioning in production?

The decision engine runs the full pipeline — eligibility evaluation, ML scoring, contact policy checks, and multi-objective arbitration — in a single real-time API call. An open benchmark fixture ships with the platform so you can measure latency on your own infrastructure, where it actually matters. Redis-cached enrichment keeps hot paths fast.

Playground liveApache 2.0Open source

Every customer.
The right decision.
Delivered in real time.

KaireonAI is the open-source next-best-action platform: visual decision pipelines, nine built-in ML engines, and an agentic AI layer that tunes itself — with your approval.

Try the Playground Self-host in 2 minutes

5,000 free lifetime decisions · no email · no credit card

studio.kaireonai.com/studio/decision-flows

Live

POST /v1/recommend

running…

Inventory15 offers

Top Offerscustomer_id: 8472

1Gold Credit Card0.94
2Premium Checking0.81
3Auto Refi · 4.25%0.67

0+MCP Tools

0Scoring Engines

0Flow Node Types

0+Connectors

Your customers expect every interaction to feel personal.

Most brands still spray and pray.

Heavyweight Suites

High licensing cost
Long implementation cycles
Limited ability to self-host

Point Solutions

One channel at a time
No unified scoring
Brittle integrations

Homegrown Rules

No ML, no learning
Hard to scale
Limited auditability

There has to be a better way.

The agentic layer

A platform that tunes itself.
Never behind your back.

Most platforms bolt on a chatbot. KaireonAI closes the loop: observe, propose, approve, apply — with the same four-eyes governance a human change would go through. The decision path stays deterministic; the configuration becomes self-driving.

AUTOPILOT

It proposes. You approve.

A background sweep watches your experiments and models. When a challenger beats the champion with statistical significance — or a model drifts — the Autopilot writes a proposal to the Recommendation Inbox. You choose the autonomy: suggest-only, four-eyes approval, or fully automatic with a complete audit trail.

evaluateAutoPromote · Bonferroni-corrected

SENTINEL

Silence is a failure mode.

A mis-scoped policy or a broken rule doesn't throw errors — it quietly stops decisions from going out. The Sentinel watches suppression and empty-decision rates every 30 minutes, raises alerts on breaches, and — only if you opt in — pauses affected flows before the damage compounds.

suppression_rate · empty_candidate_rate

GOVERNED ASSISTANT

AI that asks first.

The built-in assistant manages your whole platform in natural language — but every write it attempts returns a preview card. Nothing executes until you click Approve. Enforced at the tool layer, not by prompt engineering, and audit-logged end to end.

preview → approve → execute

MCP NATIVE

Your agents, first-class.

Point Claude Code, Cursor, or any MCP client at 160+ tools — locally over stdio or remotely over plain HTTPS. Calls are tenant-pinned, rate-limited, and audit-logged, and mutating playbooks queue for approval instead of writing directly. Agent access without agent risk.

POST /api/v1/mcp · tenant-pinned

The autonomy dial — per tenant, one setting:

suggest→auto + approval→auto

Product tour

See KaireonAI in 30 seconds

Data, decisioning, dashboards, AI assistant, and the Recommend API — end to end.

~30s narrated tour. Best viewed full-screen.

Why KaireonAI

Open source from day one. Self-hostable, production-ready, no license fee.

license cost, self-hosted

Apache 2.0 licensed. Run the full platform — visual pipelines, scoring, arbitration, dashboards — on your own infrastructure, at any scale, with no license fee.

ML scoring engines built-in

From simple scorecards to Thompson sampling bandits. Multi-objective arbitration balances revenue, margin, propensity, and engagement — not just click-through.

Real-time

single-call decision engine

The full pipeline — qualification, scoring, ranking, and contact policies — runs in a single API call, engineered for interactive, customer-facing channels. An open benchmark fixture ships with the platform so you can measure it on your own hardware.

Open Source, No License Fee

Apache 2.0 licensed. The full platform — visual pipelines, 9 scoring algorithms, 50+ connectors, 160+ MCP tools — runs on your infrastructure, at any scale. Docker Compose up in 2 minutes.

Visual Pipeline with Full Decision Trace

16 composable node types across 3 phases (Narrow, Score & Rank, Output). Every decision is forensically replayable — every rule pass/fail, every model score, every policy check is captured and queryable.

AI-Native from Day One

160+ MCP tools for Claude Code and Cursor. 90+ built-in AI assistant tools. Build and maintain decisioning config in natural language.

See It Work

Four live animations of the platform's core mechanics — the scoring math, the runtime, the learning loop, and the AI-native build flow.

Scoring · PRIE composite

How a candidate's score is composed — four weighted dimensions multiplied through a geometric mean.

PRIE composition

Geometric mean of four weighted dimensions

Live

PPropensityModel-predicted conversion likelihood

0.72 · w=0.4

RRelevanceChannel match + recency boost

0.65 · w=0.2

IImpactBusiness value + margin composite

0.84 · w=0.3

EEmphasisEditorial priority weight

0.55 · w=0.1

Composite Score0.00

score = P^0.4 × R^0.2 × I^0.3 × E^0.1 · weights configurable per RankingProfile

Runtime · real-time

The Recommend API call — typed payload to ranked decisions with full trace, end to end.

POST /api/v1/recommend

Response

—

Learning loop · Wilson 1927

The closed loop: every recorded outcome tightens the posterior. When the credible interval is narrow enough, the offer matures.

Model Health · offer_credit_card

Posterior tightens as outcomes accumulate

Immature

95% credible interval[0.20, 0.66] · width 0.46

0.0p̂ = 0.421.0

Accepted

Declined

0.34

Exposure

Width ≤ 0.20 at 95% CI = mature · exposure goes to 1.0

AI-native · MCP tools

Build decisioning by chat. Natural language picks the right MCP tool, previews the change, applies on approval.

Claude Code · MCPkaireonai · 160+ tools

⌘K

160+ MCP tools · build by chat · approve every change before it lands

Everything You Need

From data ingestion to real-time decisioning — one platform.

Decision Engine

Visual Decision Flows

Build your entire NBA pipeline visually. Each request flows through inventory, qualification, contact policy, scoring, and ranking — all configurable without code.

16 composable Decision Flow node types
9 ML scoring models (Scorecard → Thompson Bandit)
PRIE multi-objective scoring formula
Real-time single-call pipeline

AI + MCP

160+ MCP Tools + AI Assistant

Manage your entire platform from Claude Code, Cursor, or the built-in chat. Create offers, run recommendations, and analyze performance — all through natural language.

160+ MCP tools for AI IDE integration
90+ built-in AI assistant tools
Preview → approve → execute workflow
Context-aware across all modules

Intelligence

Eligibility Debug & Dashboards

See exactly why an offer was or wasn't shown. Pipeline funnel visualization, per-rule pass/fail, real-time dashboards with period filters and dimension breakdowns.

Unified customer view with eligibility debug
7 dashboards (Business, Ops, Model, Data, Attribution, Executive, Decision Health)
Decision traces with full pipeline replay
Score distribution by channel, offer, category

Data Platform

Connect Any Data Source

Bring customer data from any source. Define schemas that create real PostgreSQL tables, enrich at decision time, and segment customers with SQL-backed views.

50+ connectors: object stores, data warehouses, streaming platforms, CRMs, and more
Schema-driven: define fields, get real tables
Multi-row enrichment at decision time
Segment builder with SQL-backed views

Experimentation

A/B Testing with Uplift Measurement

Run experiments with holdout groups, measure uplift with statistical rigor, and promote winners automatically. Supports epsilon-greedy, Thompson sampling, and multi-armed bandit strategies.

Holdout groups & control variants
Z-test uplift with statistical rigor
Epsilon-greedy & Thompson sampling
Auto-promote winning variants

Enterprise Security

AES-256-GCM encryption at rest
Row-Level Security (auto-enabled)
RBAC + audit trail with hash chain
CSRF, SSRF, rate limiting built in

Self-Host Anywhere

Apache-2.0 — fully open source
Docker Compose: up in 2 minutes
Helm chart for Kubernetes
Your data, your infrastructure

Observability

Decision traces with full pipeline replay
Prometheus metrics + health endpoints
Circuit breakers + dead letter queue
GDPR/DSAR export and erasure

Governance & SDLC

Version history with publish/draft workflow
Decision flow rollback
Scheduled deployments
Change request workflows

New · Commerce

One-click personalization for Shopify

Install on a Shopify store and KaireonAI provisions the product catalog, categories, per-customer sync, and a personalized decision flow automatically. Each shopper’s real order history feeds behavioral tiers — spend, recency, frequency — that an adaptive model uses to rank offers per customer across the product page, cart, and thank-you surfaces. It works on privacy-preserving tiers, without requiring customer PII.

Explore Shopify personalization Read how it works