I Tried GLM 5.2 Inside Claude (it's f*cking incredible)
A 13-minute breakdown of the Chinese open-source model that nearly matches Opus 4.8 intelligence at one-fifth the price, and the four-step setup to wire it into Claude Code.
June 23rdAn 18-minute build diary of a custom personal OS: Telegram voice notes, live net worth, habit tracker, CRM, and the Supabase memory layer that ties it all together.
The real value of a custom personal OS isn't the dashboard interface but the portable memory layer stored in Supabase that trains AI on your patterns and can be plugged into any LLM to deliver personalized advice.
A custom personal operating system built around a portable memory layer beats any off-the-shelf SaaS dashboard because the data, not the interface, is the real asset. The build stacks a Supabase backend as the brain, a Claude-coded front end as the interface, and a Telegram bot wired to Whisper transcription so voice notes anywhere auto-classify into tasks, CRM entries, journal logs, or nutrition records without manual entry. Daily habits, starred priorities, live net worth pulled from a Google Sheet, Google Calendar, and goal tracking all surface on one screen. The actionable move is to design the schema first, treat the dashboard as disposable, and start using the system immediately so bugs and missing features reveal themselves through real daily use.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →
Pain story: disorganization cost him a missed Anthropic investment round. Introduces OS concept and credentials.

Voice note to Telegram to Whisper to Claude categorizes to Supabase to dashboard. Finance pulse via Google Sheets API.

Design mockups in Claude design, export to Claude Code, Supabase memory, Anthropic API key, write schema, Telegram bot via Botfather, security, then components iteratively.

Finance pulse, key tasks (3-5 starred each morning), daily habits with sub-tasks, creative/community/finance/wind-down routines, evening journal via mic.

Dashboard is replaceable; Supabase memory layer is the real asset. Transportable into any LLM. Daily journaling trains the AI on your patterns.

CRM: star tasks, drag to archive, Kanban view. Brain: category cards with AI-generated summaries and risk flags.

Three-monitor setup. Dashboard left, work center, Claude and GPT right. Voice to Telegram auto-populates CRM. Desktop Whisper and mobile PWA.

Free prompt PDF in newsletter (aiedgehq.co). Subscribe for follow-up on finance section.
Miles Deutscher replaced scattered SaaS subscriptions with a Claude Code and Supabase personal operating system that captures voice notes, tracks finances, manages contacts, and learns his patterns over time.
“I lost over $1,200,000 in opportunity cost because I just wasn't organized.”
“The real source here is the back end memory system, and then you can just apply this to any front end you want in the future.”
“The AI can basically act as a psychologist or a mentor and coach you through solving some of these patterns.”
“I truly believe we're moving away from the SaaS era into an era of purpose built tools.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
Miles Deutscher opens with a dollar-amount-sized mistake: a missed Anthropic investment round he never acted on because his task management was broken. Two weeks and one custom OS later, he is here to show you the fix he built with Claude Code and a Supabase backend, and to argue that the SaaS era is quietly ending.
Ordered build sequence from design through deployment for a self-hosted personal OS.
Separates the UI from the data layer. The memory layer is the durable asset; the dashboard is just skin.
“Click the link in the description, become a member of the AI Edge newsletter, join the Instagram community, find the full prompt PDF in the pinned drive.”
Soft lead-magnet CTA via newsletter + Instagram community. Prompt PDF is the hook. Mentioned twice (mid-video ~4:00 and outro). No hard sell.
00:01
00:17
00:30
00:43
00:55
01:18
01:26
01:35
01:52
02:07
02:22
02:30
02:46
03:00
03:13
03:26
03:42
03:51
04:07
04:26
04:34
04:46
05:04
05:16
05:31
05:52
05:54
06:11
06:27
06:39
06:57
07:05
07:22
07:41
07:48
08:03
08:16
08:34
08:43
08:57
09:10
09:27
09:36
09:51
10:00
10:12
10:35
10:42
10:55
11:13
11:26
11:40
11:53
12:01
12:18
12:31
12:44
12:56
13:12
13:29
13:41
13:57
14:14
14:23
14:32
14:44
15:08
15:11
15:31
15:48
15:58
16:05
16:25
16:37
16:53
17:06
17:20
17:33
17:47
17:56A 13-minute breakdown of the Chinese open-source model that nearly matches Opus 4.8 intelligence at one-fifth the price, and the four-step setup to wire it into Claude Code.
June 23rdA 20-minute breakdown of the 10/80/10 system and loop engineering — the cost-efficient way to run the most expensive AI model on the market.
June 11thClaude's top-tier "Fable" model comes off subscriptions this week — here's the one prompt that captures its thinking before it's gone.
July 6thA 21-minute first-hours take on the public release of the Mythos-class model — what it does, what it costs, and a practical framework for deploying it without burning your token budget.
June 9thA 16-minute walkthrough of the B.U.I.L.D. Framework — five steps for turning Claude Code into a system that ingests your own data, runs recurring improvement loops, and gets smarter every week.
June 28thA 9-minute screen demo of /power-up and /insights, two Claude Code slash commands that most users have never touched.
April 3rd