Chief of Staff plans
Tell it what you want in one plain sentence. It asks if something is unclear, splits the work into tickets, sends each one to the right project, and writes down every decision.
For founders who don't write code
Run a team of AI agents
from one screen.
Helme gives every project its own AI agent, puts a Chief of Staff in charge of all of them, and has a Reviewer check the work before it ships. You make the decisions. The agents do the rest.
⋮⋮ Chief of Staff
you › Ship the new checkout on shop-web this week, and fix the crash on the mobile app login. ● Chief of Staff Two tickets created and sent: ✓ T-0112 shop-web new checkout page ✓ T-0113 mobile-app login crash on Android Agents are on it. I'll tell you when the Reviewer has checked both. ● Chief of Staff One thing needs you: the payment provider asks for your approval. Added to your list. you ›
Illustration with fictional projects. Helme runs on your own Windows PC.
Built and shipped with Helme
- Wispra
- Lenvid
- Revova
- Timio
6agents side by side in one tab
1Chief of Staff who knows every project
0lines of code you write yourself
06:30morning report, every day
How it works
A small company of agents.
You are the CEO.
One agent per project
Every project gets its own agent in its own terminal pane, up to six per tab. Each works on its own branch, never on main, and reports back with evidence.
Reviewer checks first
Before anything is merged, a Reviewer agent reads the change and writes a verdict. Nothing reaches main without your yes.
shop-web
● Editing checkout/page.tsx
+ Order summary card
✓ 42 tests passedmobile-app
● Reproducing login crash
Found: null token on resume
✓ Fix on helme/login-crashbooking-api
● Rate limit on /search
✓ Build passed
→ Sent to Reviewerdocs-site
● Writing the setup guide 3 of 5 sections done
analytics
● New weekly chart
✓ Preview deployedReview
Verdict: PASS
booking-api rate limit
No secrets, tests greenFeatures
Everything a busy owner needs.
Nothing they have to type twice.
Give work in one sentence
Type the way you talk, in any language. The Chief of Staff turns it into tickets with the context, the reason, the expected result and what must not be touched, then hands them to agents that are already open.
Add Apple sign-in to the mobile app.
T-0114 created for mobile-app
Status: Approved → sent to the open agent
Status: Approved → sent to the open agent
One list of what waits for you
Every decision that needs you lands in a single list, each with its own step-by-step how-to. Answer it and it disappears. No hunting through chats and folders.
A morning report at 06:30
Wake up to one page: what shipped yesterday, what the Reviewer said, what is blocked, and what costs were used. Read it with your coffee, decide, move on.
Morning report · Saturday
- ✓ 5 tasks done, 4 reviewed PASS
- ! 1 blocked: waiting for your login
- ◷ Night shift: 3 projects, 2h 14m
- ◔ Usage this week: 41%
A night shift while you sleep
A schedule you can edit by hand: agents work at 23:00, the Reviewer checks at 02:00, lessons are written at 05:00. If the PC was asleep, the work runs as soon as it wakes.
23:00Night work on every project with a task02:00Night review, one verdict per project03:00Email scan, read only05:00Scores, costs and lessons learned06:30Morning report on your screenSafety boundaries by default
Agents never touch money, keys and passwords, real customer data, or messages to real people without your yes. They never push to main. When they hit a boundary, they stop and ask, then carry on with other work.
✕Payments, prices, ad budgets
✕API keys, passwords, .env files
✕Changing real data in a database
✕Emails and posts to real people
✕Pushing or merging to main
✓Everything else: done, checked, reported
Made for non-coders
You decide.
Agents do the work.
No code to read
Reports come in plain language, in your language. Code and commits stay in English for the machines.
Trust grows step by step
Say "next time you can decide this yourself" and the agent adds that one kind of task to what it may do alone.
Everything on your disk
Projects, reports and decisions are plain text files on your PC. No cloud database, nothing locked away.
FAQ
Questions
What exactly is Helme?
A local app for Windows that opens real terminals for AI coding agents, one per project, and arranges them in tabs and panes. A Chief of Staff agent coordinates them, a Reviewer agent checks their work, and a report area shows what is running and what waits for you.
Do I need to know how to code?
No. Helme was built for an owner who runs several products and does not write code. You give work in plain sentences and answer the decisions that need you.
Which AI does it use?
Project agents run Claude Code by default. The Reviewer is a separate agent with its own instructions, and it can be set to a model from another provider, so the code is checked by a second pair of eyes.
Can an agent spend money or leak my keys?
Not without you. Money, secrets, real data and messages to real people are blocked by default. The agent writes down what it needs and why, and waits for your answer.
Does it run in the cloud?
No. Helme runs on your own PC with Node.js, real terminals and plain files. Your code goes only where your projects already live, such as your own GitHub.
Can I download it?
Not yet. Helme is in private use today. This page only introduces it.
Stop being the bottleneck.
One screen. Every project moving. Only the decisions left for you.