Local-first workflow engine for software engineering.
It runs AI agents (Claude, Cursor) through structured, repeatable workflows against your tickets, pull requests, and documentation — with a human in the loop wherever you want one.
Install the engine
npm install -g @knoflows/engineWhat it does
Runs on your machine
It runs entirely on your own machine. Your code, tickets, and credentials stay local to the Engine process; only what a workflow explicitly sends to an AI provider or to Jira, Bitbucket, GitHub, and similar services ever leaves it.
Workflows you define
A workflow is a stored definition the engine runs. Public steps include agent, interaction, shell, git, and provider action, so a run can call an agent, pause for a person, run a command, use git, or call a provider.
Tickets, pull requests, and documentation
Subject providers supply the items a workflow runs against: tickets from Jira, GitHub Issues, or Azure DevOps Work Items; pull requests from Bitbucket, GitHub, or Azure DevOps; and documentation from Confluence. Each kind also has a manual provider.
Saved views
A view is a saved way to launch one workflow. When the workflow has a subject, the view stores the provider instance and the query. Run on all matching launches against what that query resolves now.
Terminal and MCP
The kno command browses views, launches runs, and manages jobs against an engine that is already running. kno mcp exposes those views and jobs as tools to Claude Desktop and Cursor.
Alpha / beta
Knoflows is currently in alpha/beta — actively developed, genuinely usable, but still rough in places. Expect occasional bugs and breaking changes between releases. The unsuffixed npm tag is the more stable channel. @beta has been verified by the maintainer and is not yet graduated. @alpha is freshly implemented and not verified.