one terminal for every agent you run
harnx is a native workspace for the way coding actually happens now — built around the idea that you'll have several AI agents in flight at once and you need somewhere to put them. Spawn them, watch them, steer them, ship what worked.
It's the surface, not the shell. The keystrokes you'd normally type across half a dozen terminals collapse into one window that's quiet when it should be and loud the moment something needs you.
From the first prompt to the final commit
Three motions. The same three you already do — just compressed into one place that knows how to keep up.
Open it
Point harnx at a project and the window becomes the place where the work happens. Everything you'd normally have scattered across tabs, apps, and side-windows lives inside one surface you can scan in a glance. Nothing to configure on day one — your existing repo, your existing tools, your existing shell.
Run the work
Launch as many parallel streams as the problem deserves. The environment scales with the way you actually think about a task — branching when you need to explore, converging when you need to ship, holding the context for each one separately so threads never get crossed.
Stay in flow
Outcomes surface where you're already looking. Things that need attention announce themselves; things that don't stay quiet. When you have to switch tasks, the previous one is one keystroke away — exactly as you left it, including everything the agent had said up to that point.
The number of things running for you keeps going up
A year ago you ran one agent and watched it. Now you run four and hope. Soon it'll be ten, then more — and the bottleneck stops being the model and starts being you, your attention, the shape of your desk.
The terminal was built for one human typing one command. The IDE was built for one human reading one file. Neither was built for a human conducting a small orchestra of things that all want to talk at once.
harnx is the room you put them all in. Not a model. Not a wrapper. Not a chat. A surface that assumes you've already stopped doing one thing at a time, and arranges itself accordingly.
Built for the way you actually work
Not personas in a deck. Modes you slip between in a single afternoon. harnx fits each one without making you reconfigure.
The shipper
You move fast and you move often. You want the gap between "this could work" and "this is merged" to be measured in minutes, not meetings.
The juggler
Three branches, two agents, one customer waiting. You need somewhere that holds all of them at once without any of them quietly going stale.
The explorer
You try four versions of a thing before you commit to one. You want all four breathing at the same time, easy to compare, easy to kill.
The operator
You run things that run other things. Long jobs, watchful loops, the kind of work that you check on rather than babysit.
The reviewer
Code you wrote, code an agent wrote, code somebody else wrote — you need to feel the shape of it quickly and decide.
The finisher
Eighty percent there is not done. You want a workspace that helps you carry the last twenty percent over the line without losing the thread.
What you put down. What you pick up.
Most of the friction in your day isn't the work itself — it's the scaffolding around it. harnx removes a category of small chores you didn't realize you were doing.
Stop
- Counting open terminal tabs
- Tabbing between windows to remember who said what
- Losing the thread when one agent finishes mid-conversation
- Copy-pasting context from one place to another
- Wondering if the long-running thing is still running
- Choosing between "stay focused" and "stay aware"
Start
- Letting more than one thing happen at the same time
- Treating attention like the scarce resource it is
- Spawning a thread the way you'd open a new thought
- Hearing only what's actually changed
- Ending the day knowing what shipped and what didn't
- Trusting the surface to keep up
Free for one project. $3 for all of them
No tiers to compare, no feature matrix. AI runs on your own keys — direct to your provider, zero markup — or through the Claude subscription you already have. We never proxy or meter your inference.
Free
The whole workspace, for one project.
- 1 project
- Unlimited panes, splits, and parallel agent sessions
- In-app AI agent with project + multi-pane awareness
- AI commit messages, status digest, merge-conflict resolver
- Auto-checkpoints on every agent turn, with diff + restore
- Task queue — one git worktree per task, merge back when done
- Approval queue for every agent waiting on you
- Agent loops: profiles, kanban, recurring + comparison runs
- Voice dictation — on-device and offline, or via your own key
- Self-hosted cloud: run panes + agents sandboxed on your own VPS
- AI on your own keys — zero markup, never proxied
Unlimited
Every feature, every project, one small flat price.
- Everything in Free, plus:
- Unlimited projects
- Every repo, worktree, and remote you touch in one window
- 3-day free trial
Things people ask first
Short answers on purpose. The longer ones come once you're inside.
What about my existing setup?
It stays. harnx is additive — your editor, your shell, your scripts all keep working. The window just becomes a better place to do all of it from.
Which agents does it support?
The ones you already use, and the next ones you try. harnx doesn't pick winners — it gives them all the same room.
Is this a chat app?
No. Chat is one shape work takes. There are others, and most of them aren't conversational. harnx holds the whole set.
Will I have to learn a new way to work?
Not on day one. The things you already do feel faster. The things you couldn't do before become available when you're ready.
Is there a free plan?
Yes — harnx is free for one project, with every feature included. When you want more than one project, it's $3/month or $25/year for unlimited. That's the whole pricing page.
Do I need my own API key?
For the built-in AI features (commit messages, digest, merge resolution, agent chat), yes — bring a key from Anthropic, OpenAI, OpenRouter, Groq, or Google, and calls go straight to your provider with zero markup. Agents like Claude Code just use whatever subscription or key they already have. harnx never proxies or meters your inference.
Where does my code run?
On your machine. Optionally also on your own VPS — harnx can run terminals and agents on a server you control, sandboxed, connected over SSH or an encrypted tunnel. There is no harnx-hosted compute.
One window. Every agent.
The first time you have everything running in the same place, you'll wonder how you ever did it the other way.