Corey Ganim's new roundup proposes eight services, each built around a GitHub project rather than a custom app from scratch. The strongest part of the idea is to sell an outcome a business already wants: a searchable archive, a sales-call brief, a dashboard, or clearer attribution for a mailer. The weak part is the phrase free repo when it is used as a substitute for delivery economics. The code is only one component of a service that also needs permissions, configuration, review, maintenance, and customers.
Pick the offer that matches a buyer you can actually reach, then test one small deliverable. Corey favors Agent Reach for public buying-signal alerts; my lower-friction first pilots would be a Metabase dashboard from approved sample data or a Dub campaign-tracking test. None of the video's prices or client counts are verified sales results. The eight repositories have materially different licenses and operating costs.
Watch Corey's Walkthrough
Corey describes the buyers, proposed fees and outreach hooks. The comparison below adds a check against the projects' own documentation. Treg is his separate pay-per-call lead-research tool, not a ninth project in the list.
Eight Repos, Eight Proposed Services
The fees are Corey's suggested offer prices, not measured market rates or our recommendations. The builder notes are our practical checks. No star counts are included: they change daily and do not establish that a project is suitable for a client.
| Repo | What it solves | Corey's buyer and offer | Builder note |
|---|---|---|---|
| VoiceStudio | Local voice creation and video dubbing. | A consented creator's Spanish-language videos; $1,000 setup + $750/month for up to eight videos. | The app is AGPL, but default model weights have separate non-commercial terms. Verify the selected model and every voice permission. Translation, editing, captions and publishing are still work. |
| Cloudflare Security Audit skill | A structured, six-phase source-code review with candidate and record verification. | An AI-built-app security review; $1,500 per audit + $300/month for new-release checks. | A skill is not a complete penetration test. Agree scope and authorization first; validate findings and exclusions with a qualified human before making a security claim. |
| Knowledge Work Plugins | Role-specific Claude workflows for sales, support, finance and more. | A sales-team call-prep and pipeline plugin; $2,500 setup + $500/month. | The repo is Apache-2.0, but client Claude access, CRM connectors, permission design and output review are separate. Measure time saved against the current sales process. |
| Agent Reach | Read and search public-platform content through one agent CLI. | A daily list of relevant public discussions and drafted replies; $500 setup + $400/month. | The project is read-oriented, not an automatic outreach engine. Channels and logins vary, break, or carry platform restrictions; verify each lead and approve each reply. Its official guide warns against an unrelated PyPI package. |
| Metabase | Dashboards and queries across connected business data. | A five-to-ten-metric owner dashboard and Monday summary; $2,500 setup + $400/month. | Connectors and data definitions are the hard part. The repo includes AGPL open-source and commercially licensed editions; test the edition and hosting cost that actually fit the client. |
| Paperless-ngx | Scan, OCR, tag and search business documents. | A searchable archive and proposed Claude Q&A layer; $2,000 setup + $200/month, scanning extra. | Paperless-ngx does not itself make every record accurate or add a ready-made Claude Q&A system. Plan document permissions, OCR checks, retention, backup restoration and the separate AI integration. |
| Invoice Ninja | Invoices, quotes, payment links and receivables workflows. | Reminder sequences and weekly overdue-invoice report; $1,000 setup + $300/month. | The code is source-available under the Elastic License, not unrestricted open source. White-label use has separate terms; payment processing, email and accounting review add cost. |
| Dub | Short links, QR codes and conversion attribution. | Track mailers, yard signs and vehicle wraps; $500 setup + $250/month. | A scan or click is not a booked job. Connect links to a consented conversion event and a baseline. Dub's core is AGPL with separately licensed enterprise code; hosted and self-hosted costs differ. |
Which Repo Should You Try First?
Start with a bottleneck you have observed, not the repo with the most stars or the largest projected retainer.
| If the buyer loses time on... | Start with... | First proof to collect |
|---|---|---|
| Preparing every sales call | Knowledge Work Plugins | Compare five human-reviewed call briefs with the current prep process. |
| Finding the right number on Monday | Metabase | Reconcile five dashboard metrics with their source systems. |
| Searching old paper records | Paperless-ngx | Index a consented sample and time ten real retrieval tasks. |
| Chasing overdue invoices | Invoice Ninja | Review one reminder sequence and track the actual payment date. |
| Attributing offline campaigns | Dub | Give two placements distinct links and verify a real enquiry event. |
| Monitoring public demand signals | Agent Reach | Manually check a week's findings for relevance and false positives. |
| Reviewing an AI-built app's code | Cloudflare Security Audit skill | Review one authorized repo and verify each reported issue by hand. |
| Localizing an existing video library | VoiceStudio | With written consent and a cleared model, quality-check one short dub. |
Free Code Is Not a Free Service
Four of these offers touch consequential data or rights. Voice cloning needs the speaker's permission and a model license that allows the proposed commercial output. Security work needs a signed scope and a human who can distinguish a real exploit path from a plausible-sounding false positive. A document archive needs recoverable backups and careful access controls; adding Q&A over its contents introduces another data-processing boundary. Invoicing and analytics need reconciled numbers, approved messages and a clear owner for mistakes. These checks are part of the product, not optional aftercare.
Corey's running-cost estimates cover selected servers or model calls, not the whole operation. They do not include prospecting time, implementation, scanning labor, connector maintenance, translation review, customer support, payment fees, security work, legal review or taxes. For a repeatable service, measure cost per accepted client deliverable after those inputs.
Find a Buyer Before You Build the Stack
Corey uses Treg to look for creators, hiring signals and company contacts. Its current site describes one key for pay-per-call data tools. A signal is a reason to ask a question, not proof of budget or permission to access systems. Start with a buyer interview about the last time the problem occurred, the workaround, the measurable cost and who signs off. Then offer a preview built from public, synthetic or explicitly authorized material. Do not scrape private content, clone a stranger's voice or test a live app to create a sales sample.
For Agent Reach specifically, the strongest deliverable may be a weekly human-curated relevance report, not five automated replies a day. Public posts often describe a problem without purchase intent. Label uncertain matches, provide links back to the original posts, and let the client decide whether and how to respond.
The Video's Revenue Math Is a Scenario
At 22:38, Corey calculates two clients for each of his eight proposed offers: $23,000 in setup fees and $6,200 in monthly recurring fees. The arithmetic matches the prices he proposed. It is not evidence that he acquired 16 clients, delivered these services, or earned that revenue. His claim of a 99% net margin does not follow without the missing labor and operating costs. His own practical advice is better: choose one offer and learn its buyer deeply before attempting eight parallel businesses.
Copyable One-Buyer Pilot Brief
This is an original editorial prompt, not one supplied by Corey or a tested sales system. It is designed to stop the plan when buyer evidence, permission or economics are missing.
One repo, one buyer, one result
Validate the need before building a recurring service.
Video Chapters
| Time | Topic |
|---|---|
| 00:00 | The eight-repo service playbook |
| 00:24 | VoiceStudio: multilingual creator channel |
| 03:16 | Cloudflare: code-security review offer |
| 06:08 | Anthropic plugins: a sales-team workflow |
| 09:50 | Agent Reach: public buying-signal alerts |
| 12:29 | Metabase: owner dashboard |
| 14:55 | Paperless-ngx: searchable records |
| 17:27 | Invoice Ninja: accounts-receivable workflow |
| 19:55 | Dub: offline campaign attribution |
| 22:38 | Scenario math and Corey's first pick |
Source Note and Project Links
Checked 28 October 2026. Corey's video and the supplied transcript are the source for the offer structures, proposed fees, client-count math and his Agent Reach preference. We did not run the services or verify client revenue. The eight official repositories linked in the comparison table establish what the projects describe, not whether Corey's service pricing or margins work. In particular, the VoiceStudio license notice, Invoice Ninja license, and Dub's open-core terms matter before commercial delivery. Treg is an optional lead-research service, not part of the eight-repo stack.