Supabase Compute Opens Private Alpha as API and MCP Parity Demands Grow

Key Takeaways

1. Supabase Compute entered private alpha, letting users run agents and long-running tasks inside a Supabase project next to Postgres data without a second platform.
2. SaaS customers are increasingly demanding that API and MCP capabilities match, while TallyForms reports 42% of its forms are now built by AI via Tally AI and MCP.
3. A GitHub private-repo deletion exposed a fork cascade rule labeled expected behavior, underscoring the need for clear attribution and communication in audit trails.
4. Tw93 published a reusable AI coding rule set covering agent confirmation, secrets handling, parallel worktrees, review limits, and beta gating.
5. MiaoYan expanded to iPhone and iPad as a native Swift Markdown editor with one App Store purchase across devices and a shared iCloud Drive folder.
6. Stripe won a disputed chargeback for a customer who reported a charge as fraud, at a cost of $0.86, with the author saying Stripe handled nearly everything.

1. Agent Infrastructure and Product Surfaces

1. Supabase Compute is in private alpha, and the forwarded description says users can run agents and long-running tasks within a Supabase project, adjacent to Postgres data, without needing a second platform 1. Separately, Arvid Kahl said more customers are directly or indirectly requiring API and MCP parity, arguing SaaS products must present all capabilities equally to people, agents, and machines, with UI, MCP, and API fully interchangeable 2. He cited TallyForms adoption data against the view that MCP is dead as an entry point: 42% of TallyForms are now built by AI through Tally AI and MCP 3.
2. A GitHub private repository was deleted by someone outside the organization, and the behavior was classified as expected because forks of a private repo are cascade-deleted when the original is removed, according to the forwarded account; the author said this was not clearly communicated and stressed explicit ownership and communication in audit trails 1.

2. AI Coding Practice and Developer Tools

1. Tw93 compiled a reusable set of AI coding best practices: describe background and goals to the agent rather than technical details, use the best available models, have the agent confirm new features or configurations, and split work into minimum viable increments 1. The rules also require the agent to learn the author's speaking style, change code only on explicit fix or modify instructions, confirm before commit, publish, or delete, treat any pasted secret or token as leaked and rotate it immediately, and enforce mandatory rules with scripts rather than prompts alone 2. For coding, the agent should state what it will change and affect before writing, add validation scripts with output, use separate worktrees for parallel agents, and rerun full tests after merging; for debugging, find the cause first, add logs after two failed fixes, and clear context after three failed attempts; code-writing and review agents should be separate, review capped at three rounds with only new changes after the first, and dangerous changes such as deletions, payments, and releases checked personally by the author 3. For release, commits should state what was done, how it was verified, and how to roll back; external features should get a beta switch and internal or early-user testing; every production bug and user feedback item should become a test case run before release and when switching models; fixes should address similar issues together 4.
2. MiaoYan, started in 2020, now extends to iPhone and iPad as a native Swift Markdown editor, with one App Store purchase covering Mac, iPhone, and iPad, a shared iCloud Drive folder, and a Mac version that remains free and open source 1. Kaku released nine versions in October focused on remote access: running kaku remote opens the Mac terminal, including a running Claude Code session, in another device's browser, with touch, web, and remote desktop support 2.

3. Payments, Operations, and Business Signals

1. Jack Ellis said a customer falsely reported a charge to their bank as fraud, and Stripe ultimately won the case for the customer at a cost of only $0.86; he emphasized that Stripe handled nearly all of it and he only clicked a few buttons 1. He also argued that lean operations do not mean building everything in-house: even at $2,000 MRR he paid for managed services to keep all attention on the product, and would do the same starting today, with the core goal of protecting mental bandwidth 2.
2. Bazzly's MRR rose from $15,000 to $20,000 in 10 days, according to Tibo, who said he receives almost daily messages calling it the best Reddit platform and thinks it could be another $100,000-level product; he also promoted its affiliate program 1. The quoted post described Bazzly's fit rules: most projects suit shared warmed-up Reddit accounts, while supplements, gambling, trading, and NSFW are unsuitable because those subreddits delete product recommendations outright; mismatched customers get zero results and churn, and their deleted comments damage the account pool other customers rely on, so screening should follow fit rather than signup order 2. The waitlist was removed and anyone can register, with mismatched niches routed to dedicated accounts warmed up for about 10 days in their respective subreddits 3. Tibo also said AppSumo Select requires nearly $100,000 in sales over the last 30 days and that he is $20,000 short 4.

4. Health Tracking and Personal Data Stacks

1. Levelsio said his daily magnesium intake is about 200 mg, below the roughly 420 mg recommendation, so he began taking magnesium glycinate; he cited claims that about 60% of people are magnesium deficient and that deficiency may cause anxiety and affect sleep, while saying he currently has no anxiety or depression 1. The next day he reported about 30 more minutes of REM sleep but stressed it was one day of data and not a conclusion, and that he had no nightmares 2. He said all his business and health data, including calories and sleep, runs on a VPS accessed only through Termius SSH and Claude Code on the server, letting him ask whether last night's magnesium improved sleep and pull WHOOP sleep, diet records, bedroom temperature, and air quality to rule out confounders 3.

5. Disputed and Unverified Items

1. Daniel Vassallo questioned why Mbappé would let brands including Dongpeng Beverage and Mengniu constrain him and "humiliate himself," quoting a post claiming that after Mbappé mentioned a meme involving Mao Zedong, Dongpeng Beverage, Mengniu, and Milkground terminated cooperation, Xiaohongshu removed related content, TV stations were barred from airing La Liga, and La Liga president Tebas apologized, with the post saying this could cost him millions of dollars 1. A separate account reported that CCTV demanded a written apology from Real Madrid and Mbappé, China News Service published a critical article, Zhejiang Propaganda demanded La Liga discipline, and Global Times said Real Madrid could not get away with it, while Real Madrid apologized to the Chinese people on October 28; Dongpeng Beverage and Mengniu ended cooperation, Lego's mainland China site still showed his set, Hisense's Weibo comments were filtered, and Mbappé images were covered or removed from "Bushui La" ads in multiple locations 2 3 4 5 6 7 8 9 10 11 12. Overseas reaction was split: the US House Select Committee on the CCP said he should not have to apologize, a French TF1 host called Mao a dictator, the French program The Bridge apologized to Mbappé and partners over related footage, and Tebas met China's ambassador to seek a solution, while Spanish media said Real Madrid's Chinese sponsors understood the expression as personal and kept cooperation 13 14 15 16. These accounts are disputed/unverified and are presented as claims from the cited sources.