---
title: "Overboard for Mac: Scheduling Claude Code Agents With No API Key"
date: "2026-09-23"
video_id: "7688735410867965197"
tiktok_url: "https://www.tiktok.com/@everydaylentils/video/7688735410867965197"
creator: "@everydaylentils"
category: "AI"
tags: ["overboard", "claude code", "background agents", "herdr", "mac apps"]
excerpt: "Overboard is a $49 Mac app that schedules Claude Code sessions on your existing plan. What it does, what it costs, what Herdr is and where the Codex claim stands."
---
Overboard is a $49 one-time Mac app that schedules and dispatches Claude Code sessions on your own machine, using your existing Claude Max or Pro plan instead of an API key. A TikTok from @everydaylentils demos it and says it works with the "same Claude or Codex subscription." Most of the claims match the product's site, but the site never mentions Codex, and the video skips a required runtime called Herdr.

## What is Overboard?

[Overboard](https://tryoverboard.com) describes itself as a way to "Run Claude Code Agents on your Mac 24/7." The app in the video has four sections: Projects, Scheduler, Plugins and Settings.

- **Projects** shows each tracked repo with recent commit activity, items that need review and a "Morning report" summarising what shipped.
- **Scheduler** lists recurring agents with a name, model, prompt, workspace and schedule (once, daily, weekly or every N hours), plus a queue and a run history.
- **Plugins** lists installed Claude Code plugins and where each is enabled.

The scheduled agents in the demo write blog drafts, find social posts and make meme videos for the maker's own products, so the screen recording shows real use.

## Does Overboard need an API key?

No. The pricing block says "Not another subscription. Overboard runs on your Claude plan." The [download page](https://tryoverboard.com/download) requires the Claude Code CLI "installed and logged in" on a Max or Pro plan. The [terms](https://tryoverboard.com/terms) say runs happen "on your own Claude subscription" and "consume your Claude plan's limits."

The app header shows usage meters for the five-hour session and the week, with a banner when the usage data is stale.

## Does Overboard support Codex?

The site does not say so. The word "codex" does not appear on the homepage, plugin page, terms or download page. It shows up only in a blog post that lists OpenAI Codex cloud as a competitor. Every requirement on the site names Claude only.

The app in the video does show a Claude/Codex toggle and a Codex model list with GPT-6 and GPT-5.6 entries. Codex dispatch appears to exist in the build being demoed, but it is undocumented. Treat it as unconfirmed if Codex is the reason you would buy.

## What is Herdr and why does Overboard need it?

Every Overboard run opens a real Claude Code session in its own tab in [Herdr](https://herdr.dev), a separate Apache 2.0 terminal multiplexer that describes itself as "the agent runtime." Overboard bundles Herdr and re-signs it with its own Developer ID, and the download page lists it as a requirement "for scheduled and dispatched runs." The video's "take over" feature is a Herdr tab you can open and type into. An option closes the tab on a clean finish, and failed runs stay open.

## How does scheduling work when the app is closed?

Overboard uses launchd, macOS's built-in job scheduler, so "runs fire whether or not the app is open." It also has a keep-awake menu and a global queue that the maker's blog says caps concurrent runs at eight. The homepage notes that "Overboard refuses to launch an agent inside the plugin cache."

The terms flag the main risk. Scheduled and dispatched runs start in Claude Code's auto-approve mode so they can work unattended. Those sessions read and write your files, run commands, and can publish content or call connected services.

## Can you dispatch Overboard from Slack or Telegram?

Telegram, yes. The homepage says "dispatch one-offs from Telegram or Slack," but the detailed Remote dispatch section only describes Telegram ("Message your Overboard bot on Telegram"). No Slack setup is documented, though the app's dispatch bar says "Slack and Telegram land here too."

## How much does Overboard cost?

The site lists:

- $49, one time
- A 14-day full trial with no email or credit card
- Activation on up to 3 Macs
- 1 year of updates, after which "your version keeps working"
- macOS 14 or later
- A notarized DMG sold through Polar, outside the App Store

The download page offers version 1.2.1. There is also a free MIT-licensed [Claude Code plugin](https://tryoverboard.com/plugin) with a local dashboard. The paid app adds "scheduled overnight sessions, menu bar presence, and usage meters."

## Who makes Overboard?

The homepage says only "© 2026 Overboard." The blog is bylined "Mark R," and the plugin install command points to [mrieck/claude-plugins](https://github.com/mrieck/claude-plugins), "Free indie hacker tools for Claude Code." The [mrieck GitHub account](https://github.com/mrieck) belongs to Mark Rieck of Productive Mark LLC. I found no third-party reviews or discussion of Overboard.

## Overboard vs cloud agents

The maker's [comparison post](https://tryoverboard.com/blog/background-agents-compared-cloud-vms-vs-your-own-mac) groups Cursor Cloud Agents, OpenAI Codex cloud, Devin, Google Jules and GitHub Copilot's agent as one design: fresh VM, repo clone, pull request out. Its test is whether a task touches anything outside a git clone, such as logged-in browser sessions, raw video, Keychain credentials, a local database or GUI apps. If it does, a local Mac runner has the advantage.

The post also concedes three cases where cloud agents win: blast radius (a runaway agent wrecks a disposable VM, not your working tree), wide parallel fan-out, and team work that cannot depend on one person's Mac staying awake. Its competitor pricing claims are the maker's reading and are not checked here.

## Bottom line

Overboard is a $49 one-time launchd scheduler for Claude Code that runs on your existing Claude plan, opens each run in Herdr and suits tasks that need local files and logins. Codex support appears in the app but not on the site, Slack dispatch is undocumented, and unattended runs use auto-approve mode on your own machine.

## Sources

- [Overboard homepage](https://tryoverboard.com) - positioning, price and scheduler features
- [Overboard download page](https://tryoverboard.com/download) - version and requirements
- [Overboard terms](https://tryoverboard.com/terms) - plan usage, auto-approve mode and licensing
- [Overboard plugin page](https://tryoverboard.com/plugin) - the free Claude Code plugin
- [Background Agents Compared: Cloud VMs vs Your Own Mac](https://tryoverboard.com/blog/background-agents-compared-cloud-vms-vs-your-own-mac) - the maker's cloud vs local argument
- [Overboard blog](https://tryoverboard.com/blog) - all posts
- [Overboard privacy page](https://tryoverboard.com/privacy) - analytics and payment handling
- [herdr.dev](https://herdr.dev) - the Herdr agent runtime
- [mrieck/claude-plugins](https://github.com/mrieck/claude-plugins) - the plugin marketplace repo
- [github.com/mrieck](https://github.com/mrieck) - the maker's GitHub profile


## Video description and verification notes

**Four of the five product claims in this 44 second demo check out against tryoverboard.com, and one does not: the site sells Overboard as running on a Claude Max or Pro plan with no API key, and the word "Codex" appears nowhere on the homepage, plugin page, terms, or download page, only in a blog post listing competitors. Captured at 503 views, 13 likes, 6 comments, and 6 reposts, uploaded 2026-09-23. The frames show a dark Mac window titled "Overboard" with a 40px icon rail on the left (Projects, Scheduler, Plugins, Settings), a header carrying a Claude/Codex segmented toggle plus three usage meters reading "Session (5h) 3%", "Week · all models 29%", and "Week · Fable 45%", an amber banner reading "Usage is 4h old. Runs aren't gated until it refreshes." with a Refresh button, a "Live usage" switch, and an "Open Herdr ↗" button. The Scheduler pane lists four toggled-on agents ("SocialCue Run" daily at 3:00 pm, "Write a blog for trysocialcue.com" daily at 10:00 am, "Write a blog for tryoverboard.com" daily at 2:00 am, "SocialCue go do directory post to with Opus 5" daily at 4:25 am), a "NEXT 10 DAYS" calendar strip footed with "4 runs a day · nothing today", a History tab badged 139 with green checks, green paper planes, and red "launch failed" rows, and a "New agent" form with Name, Model, Prompt, Workspace (None / Shared / Per folder), and Schedule (Once / Daily / Weekly / Every N hours) fields. Two frames cut to a full-screen Herdr terminal whose title bar reads "rks-Mac-mini.local: memebrew-plugin", then "herdr --session overboard", then "120×30", showing a finished agent report with file paths under /Users/mrieck/Sites/.**

## What the app actually looks like across the 22 frames

The presenter is a talking head over the bottom two thirds of a 720x1280 vertical frame, with the app screen recording pinned above a thin amber divider and burned-in captions straddling the seam. Large gold "#1", "#2", "#3" numerals mark the three pitch beats.

Three distinct app views appear. The Projects board (frames 3 through 5) is a three-pane layout: a left "Projects" list reading "6 tracked · synced 26m ago" with seven entries (Overboard flagged "3 to review", Social Cue "24 this week · last commit today", SEO Blog Plugin "1 this week", Demo Video Maker "2 this week", MemeBrew "8 this week", StartFlow flagged "6 to review"), each with a green commit sparkline. The center pane shows the selected project with repo chips "overboard-mac overboard-plugin overboard-website", a "NEEDS YOUR ATTENTION 3" section with three Note cards each carrying a "Resolve" button and a "Resolve all" link, and a "RECENT WORK" card reading "Mac + plugin: 1.2.0 ships per-project Herdr workspaces, the Morning report and a redesigned Scheduler" with tabs Decisions 6, Now exists 2, Code 4, Flow, Summary. The right pane is "Overview / Direction" carrying a "MORNING REPORT · ALL PROJECTS" whose first line reads "Overboard 1.2.0 and SocialCue 0.32.0 both shipped today; the board is caught up with six new work cards and two flags."

The Plugins view (frame 6) is a table with columns PLUGIN, SCOPE, COMMANDS, VERSION and six installed rows: demoday (@productive-mark-dev, "not enabled anywhere", 1 command, 0.1.0, toggle off), memebrew (@memebrew-dev, off), overboard (@overboard-marketplace, "user · all projects", 2 commands, 0.1.0, on), seoblog (@productive-mark-seo, 4 commands, on), socialcue (@socialcue, 2 commands, 0.11.0, on), and swift-lsp (@claude-plugins-official, 0 commands, 1.0.0, on).

The Scheduler view carries the header meters and the three tabs "Scheduled 4", "Queue", and "History 139", plus a one-line dispatch bar reading "Dispatch a one-off task... project, time and plugins are picked for you" with the hint "Slack and Telegram land here too" and a Dispatch button on the right. History rows include "SocialCue Run · Claude · completed · 11m 15s · 4h ago", "Morning report · Claude · completed · 5m 10s · 6h ago", and two red "MemeBrew meme: coding agent give up or hack the server · launch failed · memebrew-plugin · 6h ago" entries under a red "2 failed" label. Every row has a one-line closing report in the right column, for example "The queue still shows 0 new opportunities, which is expected this early since the s...".

Two frames open the model picker. The Claude branch lists "Opus 5 · latest", "Sonnet 5 · latest", "Haiku 4.5 · latest"; the Codex branch lists "GPT-6 Astra", "GPT-6 Sol", "GPT-5.6 Sol", "GPT-5.6 Terra", "GPT-5.6 Luna" and more. Underneath, the form reads "Slash commands and plugins work. Plugins: none picked · pick", the Workspace hint reads "Runs straight in each project folder.", and a checked box reads "Close the Herdr tab on a clean finish" with the subtext "Failures stay open so you can look."

## The claims that hold up

"Uses the plan I already pay for, no API key" is on the site in three places. The pricing block reads "Not another subscription. Overboard runs on your Claude plan." The download page lists the requirement as the Claude Code CLI "installed and logged in", specifying a Max or Pro plan and no API key. The terms are the most explicit: "Overboard launches real Claude Code sessions on your machine, on your own Claude subscription, with the permissions you configure", and "Runs consume your Claude plan's limits."

"It opens a real Claude Code session" also holds, with a dependency the video never names. The site says every run gets its own tab in Herdr, a separate Apache 2.0 terminal multiplexer that Overboard bundles and re-signs with its own Developer ID. The download page lists Herdr as a requirement "for scheduled and dispatched runs". herdr.dev resolves and describes itself as "the agent runtime", advertising macOS, Linux and Windows support, 40,432 GitHub stars, and 1,046,795 installs to date. The two terminal frames in the video are Herdr panes, titled "herdr --session overboard", each ending in a "Take over" style prompt line and a status bar reading "auto mode on (shift+tab to cycle) · 1 agent".

"Weekly usage visibility" is real in the UI and barely mentioned on the site. The three header meters and the amber staleness banner are the strongest visual evidence in the video. The site's only reference is on the plugin page, which pitches the paid Mac app as adding "scheduled overnight sessions, menu bar presence, and usage meters".

"Dispatch from Slack or Telegram" is half documented. The hero line does say "Schedule repeating agents, or dispatch one-offs from Telegram or Slack", but the detailed Remote dispatch section only describes Telegram: "Message your Overboard bot on Telegram." No Slack setup, scope, or slash command is documented anywhere on the site. The app UI backs the claim with the hint text "Slack and Telegram land here too" next to the Dispatch button.

## The Codex claim the site does not support

The transcript says "Same Claude or Codex subscription, no API key, no extra bill." Searching the fetched HTML for "codex" returns zero hits on the homepage, zero on /plugin, zero on /terms, and zero on /download. The only hits are in the blog, where Codex appears as a competitor ("OpenAI Codex (cloud)") in the cloud-versus-Mac comparison post. Every requirement and license sentence on the site names Claude and only Claude.

The app UI is the only evidence for Codex support: the Claude/Codex header toggle, the GPT-6 and GPT-5.6 model list, and frame 11, where selecting the Codex tab replaces the meters with "Session: Not available" and "Weekly: resets in 4d". So Codex dispatch appears to exist in the build being demoed while the marketing site has not caught up. Anyone buying on the strength of the Codex line in this video is buying an undocumented feature.

One smaller inconsistency sits inside the app: the header meter reads "Week · Fable 45%" while the Claude model dropdown two frames later offers only Opus 5, Sonnet 5, and Haiku 4.5. The meter names a model the picker in this build does not list.

## Price, version, and who makes it

The site prices Overboard at $49, one time, with a 14 day full trial that needs no email or credit card, activation on up to 3 Macs, and 1 year of updates, after which "your version keeps working". macOS 14 or later. Distribution is a notarized DMG outside the App Store, sold through Polar as merchant of record. The download page currently offers version 1.2.1, published sha256 77a365a7173616bbd98bc83dd79f52bb0833f737efd49787897fd22fccab595d; the DMG itself returns a 25,870,196 byte file with a Last-Modified of 2026-09-22. The app in the video is one version behind that, since its own morning report says "Overboard 1.2.0 and SocialCue 0.32.0 both shipped today."

The maker is not named on the homepage, only "© 2026 Overboard". Three independent trails converge on the same person. The comparison blog post is bylined "Mark R". The plugin page points install instructions at `/plugin marketplace add mrieck/claude-plugins`, and the GitHub API confirms mrieck/claude-plugins exists, created 2026-07-30, last pushed 2026-09-22, described as "Free indie hacker tools for Claude Code" covering DemoDay, SEO Blog, Social Cue, and Overboard. The GitHub account mrieck resolves to the name Mark Rieck at company "Productive Mark LLC". The frames corroborate it: the Herdr panes show file paths under /Users/mrieck/Sites/, the plugin table lists authors @productive-mark-dev and @productive-mark-seo, and the Projects list includes "StartFlow", which matches the startflowapp.com URL on that GitHub profile. The projects in the demo are the maker's own products, so the screen recording is a dogfooding session, not a staged dataset.

## The tradeoff against cloud agents, in the maker's own words

The video's framing ("cloud agents start from zero") is argued at length in a blog post published 2026-09-24, one day after the video. Its position is narrower and more honest than the 44 second version. It names Cursor Cloud Agents, OpenAI Codex cloud, Devin, Google Jules, and GitHub Copilot's agent as the same architecture: fresh VM, repo clone, PR out. On cost it claims Devin bills per ACU of roughly fifteen minutes of agent work, Cursor's cloud runs bill on token burn, Jules gives a free daily task allowance above which you pay, and Copilot layers usage credits on the seat price. It also cites OpenAI's own rate card estimating $100 to $200 per developer per month for active Codex use. I did not independently verify any of those five vendor pricing figures against the vendors' pricing pages, so treat them as the maker's reading.

The post then concedes three cases where cloud wins outright: blast radius, since a runaway agent destroys a disposable VM instead of your working tree; wide parallel fan-out, twelve issues in twelve sandboxes; and team work that cannot depend on one person's machine staying awake. Its actual argument is a binary test, not a preference: does the task touch anything outside a git clone. Logged-in browser sessions, raw video footage, Keychain credentials, a database on a local port, and GUI apps all sit outside it. That is why the demo's scheduled slots are blog drafts, social discovery, and meme videos rather than bug fixes.

The post also names the local failure mode the video skips: a scheduler that only fires while an app or browser tab is open is not automation. Overboard's answer is launchd, so "runs fire whether or not the app is open", plus a keep-awake menu, plus a global queue that the post says caps concurrent runs at eight with priority ordering. The homepage adds a guardrail worth noting: "Overboard refuses to launch an agent inside the plugin cache." The terms name the real risk in return: scheduled and dispatched runs start in Claude Code's auto-approve mode so they can work unattended, and those sessions read and write your files, run commands, and can publish content or call connected services.

## Key Takeaways

- Overboard is $49 one time, not a subscription, with a 14 day no-email trial, activation on up to 3 Macs, 1 year of updates, macOS 14+, and a notarized DMG sold through Polar. The download page serves 1.2.1; the app in the video reports itself as 1.2.0.
- The "no API key, uses your existing plan" claim is the best documented one. The terms state runs happen on your own Claude subscription and consume your Claude plan's limits, and the download page requires the Claude Code CLI logged in on a Max or Pro plan.
- The Codex claim is not on the site. Zero occurrences of "codex" across the homepage, plugin page, terms, and download page; the only support is the Claude/Codex toggle and GPT-6 / GPT-5.6 model list visible in the app frames.
- Herdr is a hard dependency the video never mentions. It is a separate Apache 2.0 terminal multiplexer bundled and re-signed by Overboard, and herdr.dev confirms it exists as its own funded project. "Take over in Herdr" is where the takeover claim in the transcript actually lands.
- Transcription and visual gaps: the whisper transcript is accurate to the burned-in captions with no misheard words I could find, but it omits everything material that only the frames or the site carry, including the $49 price, the Herdr requirement, macOS 14+, launchd, and the auto-approve permission mode. The frames also correct the audio's shape in two places: dispatch UI shows Slack and Telegram together while the site documents only Telegram in detail, and the header meter reads "Week · Fable" while the model picker in the same build lists only Opus 5, Sonnet 5, and Haiku 4.5.
- Unverified: no third-party discussion of Overboard exists that I could find. A Hacker News Algolia search for "tryoverboard" and for "Overboard for Mac" returned 0 hits each, and three web searches surfaced no Reddit, X, or review coverage. I could not verify an official launch date, install or customer counts, the five competitor pricing figures quoted in the comparison post, whether the Codex path works in practice, whether Slack dispatch is shipped or planned, or any of the app's behavior beyond what the frames and site copy show. I did not run the app.

## Resources

- [Overboard for Mac homepage](https://tryoverboard.com) - Establishes the "Run Claude Code Agents on your Mac 24/7" positioning, the $49 one-time price, the scheduler and queue behavior, the Telegram and Slack hero line, and the plugin cache guardrail.
- [Overboard download page](https://tryoverboard.com/download) - Establishes version 1.2.1, the sha256, macOS 14+, and the exact requirement of a logged-in Claude Code CLI on a Max or Pro plan with no API key.
- [Overboard terms](https://tryoverboard.com/terms) - Establishes the auto-approve mode for unattended runs, that runs consume your Claude plan limits, the 3-Mac license, Polar as merchant of record, and the bundled Apache 2.0 Herdr binaries.
- [Overboard free plugin page](https://tryoverboard.com/plugin) - Establishes the free MIT plugin, the localhost:8787 dashboard, the "no ANTHROPIC_API_KEY" design claim, the marketplace install command naming mrieck, and the one site mention of usage meters.
- [Background Agents Compared: Cloud VMs vs Your Own Mac](https://tryoverboard.com/blog/background-agents-compared-cloud-vms-vs-your-own-mac) - The maker's full version of the video's cloud-versus-Mac argument, including the launchd claim, the eight-run concurrency cap, the three cases where cloud wins, and the competitor pricing figures I did not independently verify.
- [Overboard blog index](https://tryoverboard.com/blog) - Establishes only three posts exist, dated 2026-09-20 and 2026-09-24, which bounds how long the product has been publicly marketed.
- [Overboard privacy page](https://tryoverboard.com/privacy) - Establishes Plausible analytics, Polar payment handling, and leftover waitlist copy referring to a future launch.
- [herdr.dev](https://herdr.dev) - Confirms Herdr is a real independent Apache 2.0 agent runtime, self-reporting 40,432 GitHub stars and 1,046,795 installs, which is the runtime every Overboard run opens in.
- [github.com/mrieck/claude-plugins](https://github.com/mrieck/claude-plugins) - The marketplace repo behind the plugin install command, created 2026-07-30, covering DemoDay, SEO Blog, Social Cue, and Overboard.
- [github.com/mrieck](https://github.com/mrieck) - Ties the anonymous "© 2026 Overboard" site and the "Mark R" byline to Mark Rieck of Productive Mark LLC, matching the /Users/mrieck/Sites paths visible in the frames.
