Release Notes

Changelog

6 releases

Last updated February 20, 2026

v2.4.0
Feature #context-engine #reasoning #workflows

Context Engine v3 — Multi-Document Reasoning

Context Engine now supports multi-document reasoning chains up to 2M tokens. Pass entire codebases, contract libraries, or documentation sets and KAIRO will reason across all of them simultaneously without losing thread.

What’s new

  • Multi-document context windows up to 2M tokens
  • Cross-document reference resolution — cite exact passages from any source
  • Semantic deduplication automatically removes redundant context
  • New context.attach() API for programmatic document injection

Migration

No breaking changes. Existing workflows benefit automatically from the larger context window.

v2.3.1
Performance #latency #infrastructure

Latency improvements — p99 reduced to 18ms

Infrastructure optimization reducing p99 latency from 34ms to 18ms across all inference endpoints. No API changes.

Changes

  • Edge caching layer deployed in 12 new regions
  • Request queue optimization reduces tail latency
  • Connection pooling rewrite for Model Router
v2.3.0
Feature #workflow-builder #visual-editor #no-code

Workflow Builder — Visual Canvas Launch

The Workflow Builder is now generally available. Design multi-step AI automations on a drag-and-drop canvas without writing JSON configuration files.

Highlights

  • Drag-and-drop node canvas with real-time preview
  • 80+ pre-built action blocks covering the most common integration patterns
  • Branch logic, loops, and conditional routing — all visual
  • One-click deploy to production from within the builder
  • Version history with diff comparison
v2.2.0
Feature #model-router #cost #optimization

Model Router — Cost optimization engine

Model Router now includes automatic cost optimization. Set a quality threshold and KAIRO routes each request to the cheapest model that meets your bar.

New capabilities

  • Cost profiles: set target cost-per-request with quality floor
  • A/B routing: split traffic between models to compare output quality
  • Real-time cost dashboard with per-workflow breakdown
  • Fallback chains: define ordered fallback if preferred model is unavailable
v2.1.0
Feature #privacy #security #enterprise

Privacy Shield — Zero-knowledge architecture

Privacy Shield is now enabled by default on all Scale tier accounts. Your data is processed in isolated sandboxes and never used to train shared models.

Architecture changes

  • End-to-end encryption for all context data
  • Isolated inference environments per tenant
  • Data residency controls: choose EU, US, or APAC
  • SOC 2 Type II certification completed
  • BAA available for healthcare customers
v2.0.0
Breaking #major-release #api-v2 #migration-required

KAIRO 2.0 — Complete platform rewrite

KAIRO 2.0 is a ground-up rewrite of the platform. API v1 is deprecated and will be sunset on March 1, 2026.

Breaking changes

  • /v1/ endpoints deprecated — migrate to /v2/ by March 1, 2026
  • Authentication now uses API keys instead of OAuth tokens
  • Webhook payload format updated — see migration guide
  • SDK minimum version requirements updated

New in 2.0

  • 10× faster Context Engine with multi-document support
  • Unified API surface across all product areas
  • Real-time streaming for all inference calls
  • New TypeScript, Python, and Go SDKs

See the migration guide for step-by-step instructions.