Built on Vercel’s eve
You know how the work gets done. Build the agents that do it.
Pick something your team does over and over — triaging tickets, chasing invoices, prepping the weekly report. Describe how it should work in plain words, and harnesst builds an agent that runs it. Stack a few together and you’ve got a small team working while you do something else.
Begin →Describe the job in plain words. The assistant writes the code and opens it for your review.
The best ideas sit in a queue.
You can see exactly where an agent would save your team hours every week. But making it real means writing a spec, handing it to engineering, and waiting for a slot that might never open. By the time it ships, the process has already moved on.
harnesst takes out the hand-off. The person who actually knows the work builds the agent, and changes it the moment the work changes.
- 01
Describe
Tell harnesst what the job is, in plain words — the steps, the rules, the exceptions. Its assistant turns that into a working agent while you watch.
- 02
Check
See exactly what the agent will do before it does anything. Approve it, tweak it, or ask for a change. Nothing goes live behind your back.
- 03
Put it to work
Sign off, and the agent starts — in Slack, over email, on your site, or on a schedule you set. Change your mind tomorrow and update it in minutes.
No black box. Just a few plain parts.
An agent is made of pieces you can open and understand: what it does, the actions it can take, when it runs, and where it shows up. Change any of them yourself. When something needs real code, the assistant writes it for you.
# support-agent/
├── instructions.md — what it should do
├── tools/refund.ts — an action it can take
├── skills/triage/ — know-how it reuses
├── schedules/daily.ts — runs every morning
└── channels/slack.ts — where it works
Change it yourself
Adjust how an agent behaves without opening a ticket. The person closest to the work makes the call and ships it the same day.
Nothing goes live by surprise
Every change waits for your sign-off, and the agent stops for you on anything that matters. No unexpected refunds, no rogue emails.
It keeps its place
Agents don't lose track when something crashes or a task takes hours. They pick up where they left off and finish the job.
Wherever your team works
One agent answers in Slack, replies to email, chats on your site, and wakes up on a schedule. Build it once, use it everywhere.
It writes the hard parts
Ask for what you want — "look up the order and refund it" — and the built-in assistant handles the technical side for you.
Yours to keep
harnesst is open source and free to run on your own servers, or let us host it. Every version is saved, so you can roll back or walk away anytime.
“The best person to build the support agent is the one who’s already answered a thousand support tickets.”
The idea behind harnesst
Product & ops
Turn the process you run every week into an agent that runs it for you. Change how it works the same afternoon you think of it.
Support teams
Hand the playbook you know by heart to an agent that follows it exactly, and checks in with you on the calls that actually matter.
Engineers
Let the team build the first draft themselves. Spend your time reviewing the work instead of turning specs into boilerplate.
Same teams. Far more done.
Run it on your own hardware. Own the whole thing.
harnesst is open source. Clone the repo, run it on your own servers, and keep your agents and your data on machines you control. There’s no vendor holding the keys — read every line, change what you want, and never worry about being locked in or shut off.
Don’t want to run it yourself? We’ll host it for you. Same software either way.
View the source on GitHub →# your server, your rules
$ git clone github.com/asiraky/harnesst
$ npm install
$ npm run start
→ running at localhost, on your box
Your first agent, before the meeting ends.
Start from a template or a blank slate. Describe the job, check the work, put it live. That’s the whole loop.