Orchestration Rebuilt
for the AI Era

Prepare, review, and refine flow changes as drafts without touching production. A draft is never executed: both scheduled triggers and manual runs continue using the last published revision, so live operations stay protected while work is in progress. When the change is ready, the editor makes the next step clear: review it, then publish with confidence.

Define trigger rules with a single when expression, using the familiar conditional syntax of modern CI/CD pipelines. Every trigger type supports it, turning multi-step configuration into one readable rule. Schedule around business calendars, public holidays, and working days, then coordinate dependent flows within a defined time window.

Loop replaces ForEach and ForEachItem. Each iteration runs as an isolated sub-execution, so a runaway loop over a huge list can no longer exhaust executor memory and impact every other flow on the instance. Define the output you need once, then gather that result across every iteration in a single step.

The new Kestra interface gives developers and platform teams a polished experience building workflows. Essential context stays within reach, with less visual noise between authoring, monitoring, and troubleshooting. Move from a flow change to a live execution with confidence, while keeping the information needed to make the next decision in view.

Plugin Artifacts bring the details behind an execution into the Kestra experience. Plugins can surface their own progress, metrics, and technology-specific stages directly in the execution view, task drawer, or detail panel. See where time is spent, identify the source of a slowdown, and understand the work happening beyond the orchestration layer, without digging through raw logs.

Define an input group once at namespace level and reference it across flows with a single line. SELECT inputs now carry human-readable labels with technical values underneath, JSON inputs validate against a JSON Schema before any task runs, and a new FORM type turns complex trigger forms into a multi-step wizard.

The extension now pulls the flow schema from your connected instance, so completion reflects the plugins you actually have. Live validation, Pebble autocompletion, topology preview with live task states, and run-from-editor are all included.

Use tag-based routing to send every workload to the compute it needs, whether that is GPU capacity, a secure network, or a regional team. Reserve capacity for critical work so competing demand cannot crowd it out, while allowing unused resources to serve other teams when appropriate. Adjust allocations live as priorities shift and define your worker topology once for consistent deployment.

Four new task runners target workloads that shouldn't be containerized: GPU training on a custom AMI, licensed software bound to a machine image, anything needing direct VM control. AWS EC2 via Systems Manager, Azure VM via Run Command, Google Compute Engine as startup scripts, Huawei Cloud CCI as serverless pods. No SSH, no public IPs.

Kestra 2.0 breaks the old rule that queue and database must be one. One execution model, pluggable queue backends, and a thin adapter layer for each system. Choose JDBC, Kafka, Redis, or AMQP to match your infrastructure and scale needs without rewriting the platform. We also send only the minimal message state over the queue and keep the database as the source of truth, reducing load and simplifying operations. More flexibility. Better scale.

Task run outputs move to dedicated storage instead of living inside execution records, keeping bulky data out of your core database. Outputs also switch to a binary format, cutting storage consumption up to 40%.
That means lower database load, less storage overhead, and orchestration that stays responsive as execution volume grows to millions of runs.

NetApp ONTAP, Veeam, Pure Storage, Dell PowerStore, Ceph, Huawei Cloud, F5, and SolarWinds IPAM join VMware, Nutanix, Proxmox, Infoblox, and Netbox. Snapshot before patching, clone volumes for dev/test, provision and register infrastructure in a single flow.

kestra/kestra:*-slim ships with no bundled plugins. Enable auto-install and Kestra fetches what your flows reference on first use, caches it, and keeps editor autocomplete current. Fast first pull, nothing you don't need.