hai init Set up once per provider
Installs whatever your editor is missing (subagents, slash-commands, the MCP server) and skips what’s already there. It pulls your tickets and specs on the spot.
One AI workflow for your whole team
Your team is split across Claude Code, Copilot, Cursor and Codex. HAI sets all of them up the same way, with the same subagents and slash-commands. Then it pulls your Jira tickets and Confluence specs into plain Markdown the AI can actually read.
$
npm install -g @2amtech/hai $ npm install -g @2amtech/hai
$ hai init
Hai has initialized your AI environment!
* AI Provider: Claude Code
* Ticket Provider: Atlassian (Jira Projects: PROJ)
* Specification Provider: Atlassian (Confluence Spaces: TEAM)
Recommended workflow:
1. In your AI prompt run /pull — Sync your tickets and specs to .ai/
2. /implement TICKET-KEY — Implement a ticket assigned to you
3. Commit, push, and open a PR
Added prompt commands:
/pull — Sync tickets and specs to .ai/
/implement — Implement a ticket
…
Added subagents:
architect — Design implementation plans
researcher — Explore specs, tasks, and codebase
…
Happy coding!
Set up once, then /implement
Setup is a one-time thing. After that, you hand your AI a Jira key and it takes the ticket from plan to code.
hai pull Refresh your specs
init already pulled everything. Reach for hai pull mainly when specs change and you want the latest copy locally.
/implement JIRA-123 the main loop Hand your AI a ticket key and it runs the whole job:
- 1 Pulls the ticket over MCP
- 2 Researches the specs & code
- 3 Drafts a plan
- 4 Implements it, once you approve
What HAI wires up for you
The connective work between your project tools and your editor that you'd otherwise do by hand.
Built-in MCP server
Run hai mcp and any MCP-aware assistant can call its three tools over stdio: pull, ticket_pull and image_download.
Tickets & specs as Markdown
Jira issues and Confluence pages land in .ai/ as Markdown, with comments, links, custom fields and the page tree kept intact.
Six built-in commands
They show up as slash-commands in your editor: implement, pull, pull-tickets, pull-specs, optimize-ai and review-changes.
Specialized subagents
Seven subagents come installed, from researcher and architect to backend-dev, frontend-dev, refactorer, security-reviewer and verifier.
Parallel implementation
The implement command fetches a ticket, researches it, plans the work, then runs the domain agents at the same time and checks each one.
Recursive linked issues
hai grabs a ticket and every issue it links to in a single pass so AI has full ticket context.
Authenticated image fetch
Attachments sit behind auth, so your AI can't just fetch them. image_download pulls them with the right credentials and caches them locally.
Plain, git-ignored .ai/
It's all Markdown you can read, diff and commit. No database, nothing proprietary, and switching providers later doesn't change how you work.
Installs into any editor, pulls from Jira and Confluence
Everyone picks their own editor. Each one gets the same setup, installed where it expects to find it, so the whole team matches.
Ticket & spec sources
Tickets + Specs
Atlassian
Jira for tickets, Confluence for specs. It follows linked issues, keeps custom fields and comments, and turns Atlassian rich text into Markdown.
Tickets + Specs
Local
Reads tickets and specs from folders on disk. Good for offline work, tests, or docs that live in the repo.
No-op
None
Switches a source off. Use it when you only want the subagents and commands, with no Jira or Confluence attached.
Two commands and the whole team is set up
HAI is one npm package. Each teammate installs it and runs hai init once.
That wires up their editor and does the first sync, so there's no separate pull to run.
Apache-2.0 Node ≥ 16 MCP over stdio
1
Install globally
$ npm install -g @2amtech/hai 2
Set up and sync
$ hai init hai init runs the initial sync for you, so your .ai/ folder is
ready right away.