Hands-on experience with 🛠️ Dev tools · 🤖 AI agents · 🐧 Linux. Turning 2am projects into guides you can actually follow
- Tweets 785
- Following 102
- Followers 61
- Likes 120
10 million embeddings fit in about 4 GB of RAM with turbovec, where float32 needs 31 GB. The accuracy you trade away for that is small.
1
9
You can restrict a search to an allowed list of ids without losing accuracy, so per-user filtering is not a bolt-on.
1
5
It embeds in your app instead of running as a separate database.
botmonster.com/ai/turbovec-q…
10
Kudu is a system cleaner with no ads, no upsell and no telemetry. Its cleaning rules ship as plain JSON files you can read before you trust them.
1
10
It also scans for malware and gathers Windows privacy switches in one panel.
Same app on Windows, macOS and Linux.
Deleting system files is risky, so make a restore point first.
botmonster.com/hardware/kudu…
11
ConvertX converts files on your own server, wrapping 19 tools including FFmpeg, ImageMagick and Calibre. The first account created is admin, so never expose a fresh install.
botmonster.com/self-hosting/…
8
wigolo gives a coding agent web search with no API key and no per-query bill. Every hit comes back with a verbatim quote pinned to byte offsets in the source, so you can check the claim.
1
14
Everything it fetches is cached, so asking again is free.
One call fans out many queries across many engines.
The tools are keyless. Writing the answer still costs a model call.
botmonster.com/ai/wigolo-loc…
12
A service worker under 100 lines takes static assets from 200-500ms over the network to 1-5ms from the Cache API. Cache-Control cannot do that.
botmonster.com/web-dev/servi…
10
Claude Code's --worktree flag runs 10+ parallel agents on one repo, each scoped to its own branch with no file collisions. incident.io went to 4-7 agents per dev in four months. Cloning the repo wastes the shared history. How many agents do you run at once?
2
15
Twenty concurrent terminal sessions use roughly 7% CPU combined.
Worktrees stop filesystem clashes, not git merge conflicts; scope by file.
Past 10 agents the human merge step becomes the real bottleneck.
botmonster.com/ai/parallel-a… #ClaudeAI
26
Systemd timers beat cron on every scheduled task: journal logging, dependency ordering, cgroup limits, missed-run catch-up. Cron has run unquestioned since the 1970s with zero sandboxing. The one reason left to keep it is legacy boxes. cron or systemd timers on your boxes?
1
1
15
OnUnitActiveSec kills overlapping rsync runs, no flock wrapper.
MemoryMax and CPUQuota cap a runaway backup before it starves your DB.
RandomizedDelaySec spreads certbot renewals across servers.
botmonster.com/self-hosting/… #SelfHosted
1
9
llmfit ranks 5,372 local models against your RAM, CPU and GPU, then benchmarks the winner. On my RTX 4080 it predicted 40.9 tokens a second and measured 68.2.
botmonster.com/ai/llmfit-whi…
15
WLED flashes onto an ESP32 from a browser in 60 seconds. Add Home Assistant Assist and you have voice-controlled LED strips under $30, no cloud.
botmonster.com/smart-home/vo…
12
Meilisearch and HTMX give typo-tolerant search under 50ms in about 14 KB of JS. React plus InstantSearch ships 80-120 KB before you write a component.
botmonster.com/web-dev/full-…
18
Cobra plus Bubble Tea compiles to one static Go binary under 15MB that stays scriptable in CI and shows a full TUI when a person runs it.
botmonster.com/coding/build-…
11
One Rust codebase builds the web, desktop, mobile and server versions of an app. Dioxus is the framework doing it, and hot patching drops changed code into the app while it runs.
1
32
Web builds start near 50 KB. Portable desktop binaries stay under 3 MB, a rounding error next to an Electron app.
1
7
The 0.8 line is still alpha, so pin a version and expect churn.
botmonster.com/coding/dioxus…
9