If you run more than one coding agent at a time, you already know the tax. The tab-cycling. The “which one is stuck?” shuffle. The permission prompt that has been sitting there for ten minutes while you made coffee, and the agent that has been burning your rate limit doing nothing. Farol is a coding agent attention monitor for macOS that tries to delete that tax entirely: a terminal that tells you, at a glance, which of your agents is working, which one needs you, and which one is done.
Built by developer Lucas — the name is Portuguese for “lighthouse” — and launched on Product Hunt on 10 October 2026, Farol is free, MIT-licensed, and collects no telemetry. Our verdict up front: if you regularly run two or more Claude Code or Codex sessions in parallel, Farol solves a genuinely annoying problem with unusual elegance. But it is early software, it is Mac-only, and it asks you to move your terminal sessions into it to get the full benefit.
What this coding agent attention monitor does
The core idea is disarmingly simple. Each of your agent sessions gets a dot in Farol’s sidebar: hollow when idle, breathing while the agent works, rippling when it is waiting for you, filled when it is done. The same state surfaces in the menu bar and at the notch, so you can be in an entirely different app and still see that an agent needs you. The maker built it because he was running three or four Claude Code and Codex sessions at once and kept cycling through tabs looking for the one stuck on a permission prompt — and that origin story tells you exactly what the tool is and is not. It is not a coding agent. It is the radar screen above your coding agents.
Crucially, Farol does not wrap the agents. Claude Code and Codex run exactly as they would in any terminal, and they report their state through their own hooks — hooks you switch on with one click, according to the launch notes. That design decision matters more than it sounds: there is no proxy, no reimplementation, no session state trapped inside a proprietary wrapper. If Farol disappeared tomorrow, your agents and their transcripts are untouched. From an agency perspective, that is the kind of architecture we respect — reversible, non-invasive, honest about where it sits in the stack.
Beyond the attention layer, Farol ships a small but opinionated workflow. Start a new task by describing it — or by picking a GitHub issue or a Jira ticket — and it opens a git worktree on a fresh branch, then starts the agent with the task as its first prompt. When the agent finishes, a review view shows everything the session changed as one diff; commit, push, and open the PR from the same window. And when a rebase or merge stops on conflicts, you get a three-column merge view. If you have ever lost an afternoon to a half-finished merge across parallel agent branches, you know why that last feature is on the list.
Under the hood, the terminal itself is libghostty — the rendering engine behind Ghostty — so it renders like Ghostty and reads your existing Ghostty config. The application around it is native Swift. In practice that means fast rendering, low overhead, and a native Mac feel, which is precisely what you want from something that sits open all day.
Pricing: genuinely free
Let us get the pricing question out of the way with facts we can verify. The Product Hunt listing shows a price of 0 USD. The maker states plainly: free, MIT licensed, with no account and no telemetry. There is no paid tier to upsell you into, no per-seat billing, no “pro” badge dangling over the features you actually want. The cost of Farol is your time: installing it, enabling the hooks, and moving your agent sessions into its terminal.
The MIT licence deserves a sentence of its own. If you are evaluating tools for a team or a client engagement, MIT means you can read the source, audit exactly what the hooks report back, and fork the project if the maker moves on. For an agency like ours — nine-plus years and more than 300 shipped projects of being asked “what data does this tool touch?” by cautious clients — that auditability is worth more than any feature checkbox. No-account plus no-telemetry means your prompts, transcripts, and file contents never leave your machine by design, not by promise.
Who Farol is for — and who should skip it
Farol earns its place for a specific kind of developer: the one already running parallel agent sessions. If you keep two to four Claude Code or Codex sessions going — one refactoring, one writing tests, one researching a library — the attention layer pays for itself the first week. The permission-prompt stall is the quietest productivity leak in agentic development: the agent waits, you do not notice, and a five-minute job becomes a thirty-minute one. A ripple in the menu bar is a cheap fix for an expensive habit.
It is also a natural fit for tech leads and agency teams. The worktree-per-task model is the right isolation primitive for parallel agents — each task gets its own branch and working directory, so sessions cannot step on each other’s files — and the single-diff review view gives a lead a sane way to audit what the agents actually changed before anything merges. If your shop is building AI agent development workflows for clients, this is the category of tooling worth standardising on: the ones that make agent output reviewable instead of mysterious.
Now the honest other side. If you run one agent session at a time, Farol buys you almost nothing — a well-configured notification hook on your existing terminal covers 80 percent of the value. If you are on Windows or Linux, stop here: there is no build, and none is promised. If you are on an Intel Mac or anything below macOS 14, it will not run. And if your primary agent is Cursor rather than Claude Code or Codex, the state reporting is built on the hooks those two CLIs expose — check compatibility before you rearrange your workflow around it. You can browse how it is categorised alongside other new launches in our AI Tools Directory to see where it sits in the current wave.
Watch-outs and honest limitations
The maker is upfront that Farol is early and expects rough edges — he says he uses it as his daily terminal, which is encouraging, but “daily driver for the author” and “ready for your team” are different bars. Here is what we would flag before recommending it to a client:
- Platform lock-in is absolute. Apple Silicon plus macOS 14 or later only. A mixed-platform team cannot standardise on it.
- It is a terminal replacement, not an overlay. The full value — worktrees, review view, conflict handling — only exists if your agents run inside Farol. If you love your current terminal setup, that is a real switching cost.
- Two-agent world. State reporting is built for Claude Code and Codex. The agent ecosystem is wider than that, and it is moving fast.
- Early-stage stability. Expect the occasional rough edge; this is a weeks-old project, not a hardened product. Do not bet a production deadline on day one.
None of these are disqualifying for the tool’s actual audience — an individual developer or a small team already living in the Claude Code/Codex world on modern Macs. They are the difference between trying a promising tool with open eyes and adopting it on hype. We have shipped enough client work to know that the distinction matters.
Farol Remote: answering your agents from your phone
The most interesting follow-up arrived within days of the launch: Farol Remote, an iPhone companion app from the same maker. The pitch is the coffee-run problem made explicit — your agent asks “The fix is in. What should I do with it?” while you are away from your desk, and instead of the work stalling until you return, you answer from your phone.
The phone app lists your sessions with their status and opens each one as the agent’s conversation rather than a wall of terminal text. When the agent needs an approval or asks a question, the possible answers appear as buttons — one tap and the session carries on. You can also reply with a slash command, send a screenshot for the agent to look at, review what changed file by file, and start a new session in a project. It notifies you when an agent waits or finishes, and a lock-screen Live Activity keeps count of working versus waiting agents even with the app closed. One phone can track several Macs and servers at once, with each session labelled by its host.
This is the feature that takes Farol from “nice terminal” to “new workflow category.” Long-running agent tasks — migrations, test-suite repairs, research spikes — are exactly the jobs you want to fire off and check from the sofa. Approval-by-phone closes the loop that makes unattended agents practical. Launch details are still emerging, so treat it as a companion to evaluate alongside the Mac app rather than a standalone product.
How it compares to the alternatives
Farol did not invent the agent-attention category; it joined a genuinely crowded week for it. Here is how the field looks from what each project has publicly put forward:
- Shrine is a small macOS app that shows your agent sessions grouped by project, rendered as glowing spirits that hover while they work and dim when idle. It is free, needs no account, runs on macOS 14+, and adds a secrets proxy (“Treasure Chest”) so your keys stay in 1Password while agents see only placeholders. More ambient, less workflow.
- NapCat takes the Dynamic Island route: a free macOS app that wraps your MacBook’s notch area to show which agents are working, which need attention, and which finished. It supports Claude Code, OpenAI Codex, Cursor, and other recognised agent CLIs, and lets you jump straight back to the right session instead of hunting through terminal windows.
- CrewTower is the maximalist of the group: a native Mac notch app that keeps your prompts, files, and transcripts on your machine with no account and no analytics, and supports a long list of agents — GitHub Copilot, Aider, Cline, Windsurf, Warp, and many more — across iTerm2, Ghostty, Kitty, Alacritty, WezTerm, and tmux panes. If agent breadth is your priority, it casts the widest net.
- OpusBar lives in the menu bar as a pixel cat whose poses map to agent states, with session cards grouped under “Attention” and plan-limit tracking showing your five-hour and weekly windows for Claude Code and Codex accounts. The standout feature is cost awareness: warnings at 80 and 95 percent of your usage limits.
Farol’s position in that field is clear: it is the only one that is a full terminal rather than an overlay, the only one with the git-worktree task model baked in, and the only one with a phone companion for approvals. If you want attention monitoring without changing how you work, the notch apps are the lighter touch. If you want the whole parallel-agent workflow — isolate, monitor, review, merge — Farol is the most complete answer, at the cost of adopting its terminal.
From our experience: the right tool for the parallel-agent era
Here is the practitioner take, earned the hard way. Running coding agents in parallel is no longer an experiment; it is how a meaningful share of professional development gets done in 2026. But the tooling around it is still catching up. Most developers we talk to manage parallel sessions with a mess of terminal tabs, tmux windows, and hope — and the failure mode is always the same: an agent sits idle waiting for input, and nobody notices for twenty minutes. That is not a model-capability problem. It is an attention-routing problem, and attention routing is exactly what Farol is.
Two design choices tell us the maker has lived this problem. First, the hook-based state reporting instead of a wrapper: wrapping agents is how you get brittle integrations that break every time the CLI updates; reading the agent’s own state signals is how you get something that survives version churn. Second, worktrees per task: anyone who has run two agents in the same working directory has watched them clobber each other’s edits, and the worktree model eliminates the entire class of problem rather than papering over it. These are the choices of someone who has been burned, not someone chasing a trend.
The gotcha we would warn a client about is adoption depth. Farol’s value compounds with session count: at four parallel sessions it is indispensable; at one, it is decoration. Before rolling it out to a team, run it solo for two weeks, count how many stalls the attention layer actually catches, and then decide. That is the same advice we give for any agent workflow tooling: measure the specific friction it removes, not the demo it gives.
Frequently asked questions
Is Farol really free?
Yes. The Product Hunt listing shows a price of 0 USD, and the maker states it is free and MIT-licensed with no account and no telemetry. There is no paid tier. Because it is MIT-licensed, you can also inspect the source code to verify the no-telemetry claim yourself.
Does Farol work with Cursor or other coding agents?
Farol’s state reporting is built on the hooks exposed by Claude Code and Codex. Those are the two agents the maker names and supports. If your primary agent is Cursor, Windsurf, or another CLI, check compatibility before switching — alternatives like CrewTower or NapCat cover a broader agent list.
Does Farol run on Windows, Linux, or Intel Macs?
No. Farol requires Apple Silicon and macOS 14 or later. There is no Windows or Linux build, and the maker has not announced plans for one. It is early software, so platform support could change — but do not plan around it.
What data does Farol collect?
According to the maker: none. No account, no telemetry, and your prompts, files, and transcripts stay on your Mac. The MIT licence means the claim is auditable in the source code, which is the strongest version of a privacy promise a developer tool can make.
Why not just use tmux with notification hooks?
For a single session, that is genuinely fine — a good notify hook covers most of the value. Farol earns its keep at two or more parallel sessions, where it adds the visual state sidebar, the menu-bar/notch presence from any app, the worktree-per-task isolation, the single-diff review flow, and the phone companion. It is a workflow, not just a bell.
Farol is a small tool solving a specific, real problem — and in the current AI gold rush, that specificity is its strength. If you are a Mac-based developer already living in parallel Claude Code or Codex sessions, it is worth an afternoon of your time: free, open-source, and built by someone who clearly runs the workflow it serves. If you are evaluating agent tooling for a team or a client project and want a second opinion, talk to us — this is the layer of the stack we think about every day.