All your coding agents.
One command center.
Claude Code, Codex, Cursor and more already keep your history on your Mac and servers. Harness Bay puts every session in one inbox — then lets you continue, start, steer, and search them through each agent’s own CLI.
Illustration of the app with sample sessions.
Works with the agents you already use — and the list keeps growing
- Claude Code
- Codex
- Cursor
- Grok Build
- Kimi Code
- OpenCode
- Hermes
- ZCode
- OpenClaw
Every lab shipped a great harness. None of them can see the others.
Your real week doesn’t live in one app. It’s a Claude Code run in the terminal, a Cursor chat in the IDE, a Codex task on a dev box, an OpenClaw job on a VPS. Each tool keeps its own history, in its own format, in its own folder — and shows you only its own slice.
A lab’s own app
- Sees
- Its own agent
- History
- Its own sessions
- Acts on
- Its own sessions — locally, remotely, or in its cloud
A multi-agent launcher
- Sees
- Several agents
- History
- Mostly what it launched
- Acts on
- What it launched, in fresh workspaces
Harness Bay
- Sees
- Every supported agent — nine today, more coming
- History
- Every session already on disk, wherever it started
- Acts on
- Any of them, on your Mac or over SSH
They’re where the work happens. Harness Bay is where you see all of it.
Not another agent. The layer above them.
Harness Bay has no model and no API keys of its own, and it doesn’t ship a coding agent. It reads the history your harnesses already keep, and everything it does — Ask included — runs through the CLIs you already installed and signed in to. A session you continue is that agent’s own session: its tools, its sandbox, its plan, its bill.
So when a lab ships a new model, it usually shows up in the picker the same day — read from the agent’s own CLI. Harness Bay just keeps the whole fleet in view.
Harness Bayinbox · continue · start · steer · askSee it all. Act on any of it. Find it again.
Everything you’ve already done, already here.
Open Harness Bay and your history is waiting. It reads what each agent already stores — nothing to import, nothing to migrate, no sign-in needed just to look.
- Every session, wherever it started — the terminal, an IDE, a desktop app, or Harness Bay itself.
- Full transcripts wherever the agent keeps them — edits with diffs, commands, reasoning, not just chat bubbles.
- Your servers too. Agents on a dev box or VPS appear over SSH, grouped by machine and project.
- Monitor shows what’s running right now and what needs your attention.
Pick up any session — in the agent’s own tools.
Harness Bay drives the agent CLIs you’ve already installed. Continue a Claude Code session and it’s the same session you’d resume in Claude Code.
- Start in any folder — switched to the branch you pick — on this Mac or an SSH machine.
- Queue or Steer while it works — follow-ups wait their turn, or redirect the run mid-turn where the agent supports it.
- Stop any run you started. Where an agent hands permission requests to Harness Bay — Claude Code, Grok Build, Hermes, ZCode — Approve or Deny them inline. Others run under the mode you pick at start — except Kimi Code, whose headless runs always auto-approve.
- Pick model and effort from each agent’s lineup, read live from its CLI where it exposes one.
Cursor IDE chats continue in a linked Cursor CLI session, since the IDE doesn’t accept outside writes. Remote sessions are text-only today.
npm run test:auth -- --coverage
Ask your history. Get receipts.
You remember the problem, not which agent solved it. Ask in plain language — answers about your work cite the sessions they came from.
- One question across every agent on this Mac — “what was I working on last week?” answers from your real history.
- Choose who answers. Several agents can be the brain; every agent’s history is searchable.
- Keep what matters as editable Knowledge pages, sources attached.
- Optional digests sharpen recall. Off until you approve a cost estimate.
Ask runs through the agent you pick, so the context an answer needs goes to that provider, under its terms and your plan. SSH machines aren’t in Ask’s index yet.
The rest of the workbench.
Browser
Your dev server beside the chat. Capture the page, pick an element, or box a region, then drop it into your next message.
Terminal
A shell in the session’s local project folder — for running commands without leaving the window.
Files
Click any path an agent mentions to open it at the right line — on whichever machine the session lives.
Usage
Measured tokens and estimated cost across providers for 7, 30, or 90 days — only what each agent actually recorded.
Knowledge
Notes and saved answers, searchable alongside your history and readable by Ask.
Remote machines
Any host in your SSH config. When one is unreachable, you get the evidence and a next step — never a fake “sign in.”
No favorites. Here’s exactly what works.
Nine agents today, with more added as new harnesses ship. Deeper controls depend on what each agent’s CLI exposes, so here they are, agent by agent. In the app, unavailable controls are hidden or labeled — never faked.
Stop: a tick means the agent’s own cancel on text runs; screenshot and fallback runs end the process. Approve in Harness Bay applies to text runs; screenshot runs follow the agent’s own permission mode. Steer applies to text runs started or continued in Harness Bay. SSH runs are text-only. Ask searches every agent’s history on this Mac; “Answers Ask” is who can be the brain.
† Claude Code: on continued sessions. ‡ OpenCode: when connected to its native server. § Hermes: default-model sessions started in Harness Bay. ¶ Kimi Code: native on this Mac; SSH runs end the process. New agents land as adapters are added; this table changes with each release.
What it is. What it isn’t.
Harness Bay is
- A Mac appIt runs on your machine and talks to your SSH hosts directly.
- A layer over your agentsIt reads the history they keep and drives the CLIs you already use.
- One inbox for every sessionIncluding work you started long before you installed it.
- A way to search and citeYour past work, across every agent, with links back to the source.
- FreeYour agents bill you exactly as they do today.
Harness Bay isn’t
- An agent or a modelIt doesn’t write your code. Your agents do.
- A replacement for Claude Code, Codex, or CursorYou still install them and sign in the usual way.
- A cloud serviceNo Harness Bay account and no Harness Bay server.
- A credential storeYour sign-ins stay with each agent.
- Offline AIAsk and digests run through the agent you choose, under its terms.
- On Windows or Intel Macs — yetApple silicon only for now.
It reads every transcript you have. So here’s the deal.
Reading your agents’ history is the whole product. The posture has to be simple enough to check.
- Your data
- Stays in your agents’ own folders and on your machines. There is no Harness Bay server to send it to.
- Account
- None. Nothing to sign up for.
- Telemetry
- None — no analytics, no tracking, nothing sent to us. On its own it checks the public release feed for updates, runs your agents’ own sign-in and model-list commands, and connects to the SSH hosts you add.
- Credentials
- Sign-in runs each agent’s own login flow, and tokens stay where that agent keeps them. Harness Bay never stores them.
- AI features
- Ask and digests send the context they need through the agent you pick, as short marked sessions. Digests never spend a token until you confirm an estimate.
- Acting
- Read-only until you send something. The exact command it will run is visible at sign-in and when a session starts.
Good to know.
Do I need every supported agent?
No. Harness Bay works with whichever ones you have — even just one. Onboarding shows what it found and points to the rest if you want them.
Will it change my sessions?
Not unless you ask it to. Browsing is read-only. Harness Bay writes to an agent only when you send a message, start a session, or approve a request, and it uses that agent’s own CLI to do it. Ask and digests run as short, marked sessions in the agent you pick — Harness Bay hides them, but that agent’s own history list may show them.
Is my data sent anywhere?
Not to us — there’s no Harness Bay server. Context goes only to the agent you pick for Ask or digests, and to your own SSH hosts when you connect to them — plus whichever agent you message or start, exactly as using it directly would.
Does it work with ChatGPT or the Claude desktop chat?
No. Harness Bay is for coding agents and the history they keep on disk — not general chat apps.
How is this different from a multi-agent launcher?
Launchers are built around new sessions they start, usually in fresh workspaces. Harness Bay starts from the history you already have — every session from every agent, including ones started in a terminal, an IDE, or on another machine — and lets you act on any of them.
What does it cost?
Harness Bay is free. Running agents and using Ask spends your own plan or API usage with each provider, exactly as using those agents directly would. Usage shows what each agent recorded.
How does it reach my servers?
Over SSH, using the hosts already in your ~/.ssh/config. Remote sessions can be read, started, and continued, as text-only for now.
Bring the whole fleet into one bay.
Free for Mac. Opens to the history you already have.