Skip to main content
Yukta runs workflows: graphs of small steps that pass messages to one another. You describe a workflow as JSON, store it on the server, and start it from anywhere that can make an HTTP request. While it runs you can watch it live, pause it, skip a step, approve a decision, retry a failure or stop it.
Yukta is pre-beta (0.0.1-SNAPSHOT). The workflow JSON, the REST API, the MCP tools and the plugin contract may change before v0.1.0, and there are no tagged releases yet, so you run Yukta from source.

​
What a workflow looks like

A trigger starts the flow, processors transform, route and call out, and terminals end each branch. Processors can have named output ports such as success and error, so one node can send messages down different paths.

​
Why Yukta

Workflows are data

A workflow is a JSON graph. There is no language to learn, and a graph is easy for people, tools and AI agents to generate, diff and review.

One definition, many callers

Start the same workflow from the REST API, the web UI, the command-line client or an MCP client such as Claude.

Streaming, with back-pressure

Messages stream through the graph. A slow node slows the nodes feeding it, so memory stays bounded without queues between steps.

Observable and steerable

Live graph view, log and status streams, and run controls: pause, resume, step, breakpoints, skip, retry, restart and stop.

Failure is a first-class path

Retries with backoff, error routing to dead-letter nodes, failure budgets and five independent time limits are built in around every node.

Safe to start with

The server listens on loopback only, refuses to run operating-system commands until you allow them, and requires an API key before it accepts remote connections.

​
What it is good for

  • CI/CD and quality gates: run build and test stages, route on exit codes, stop a pipeline on failure.
  • Data pipelines: split a batch, filter, map and load, with retries and dead letters.
  • Operations automation: health checks with alert paths, scheduled jobs, fan-out notifications.
  • AI agents: model calls, tool use over MCP, per-user memory, approval steps and budgets that stop runaway loops.

​
Where to go next

Quickstart

Run the server and your first workflow in five minutes.

How Yukta works

Sessions, workflows, nodes, messages and executions.

Defining a workflow

The full JSON shape and the options every node accepts.

Plugins

Every built-in trigger, processor and terminal.

MCP server

Let an AI agent create and run workflows.

REST API

Endpoints, authentication, paging and streams.