Volodymyr Panchenko (@Portall) header
Volodymyr Panchenko (@Portall) profile picture

Portal: your own AI. Your mirror, working for you: it learns what matters to you and helps make it happen. Built DMarket, acquired by Mythical Games. Dad. 🇺🇦

Santa Monica, CA
Joined March 2018
Photo by Volodymyr Panchenko (@Portall)Photo by Volodymyr Panchenko (@Portall)Photo by Volodymyr Panchenko (@Portall)Photo by Volodymyr Panchenko (@Portall)Photo by Volodymyr Panchenko (@Portall)Photo by Volodymyr Panchenko (@Portall)Photo by Volodymyr Panchenko (@Portall)Photo by Volodymyr Panchenko (@Portall)Photo by Volodymyr Panchenko (@Portall)Photo by Volodymyr Panchenko (@Portall)
One job: run the owner's errands. Hands looks them up on its own computer and returns at most three options; it never picks. A second Bot is the checker. When a receipt lands, it recomputes the result hash against the proof card. On a match it stays quiet. On a miss or a missing file it writes one FIX line. The director speaks only on a miss or a silence.Morning re-check of 2026-09-26: this setup has been live for its owners since 2026-09-22. 8 errands closed, 0 purchases without a written pick, 0 waiting on a decision. Closed means the options file or the receipt landed. The sample run's checker re-check read match; no miss has been logged yet.x.ai/bot/8ByzhAF5ehyCC0AREHU…
We’re inviting one Grok Bot user to a Starship launch. You could win the invite by sharing the ways you've integrated @Bot into your work. The invite includes a plus one. How to submit and some examples we love:
1
113
Congrats to the @pika_labs team 🎉 Pika was in Portal's video stack in 2025, so we ran H3 Flex on the free public space: a robot walks forward and turns right — 480p, 5 s, seed 42, a 21-second generation. The best review of a model is somebody else's run; here's ours.
Meet Pika H3 Flex, debuting today as the #1 open weight model on DesignArena’s t2v board. It’s built for professional iteration, with exportable audio stems, resolutions up to 4k, and faster-than-realtime generation. It can produce a 15-second clip with soundtrack in as little as 7 seconds. So, yeah. It’s kind of a flex 💅
1
21
For anyone re-running it: huggingface.co/spaces/Pika-L… · Text · 480p · 16:9 · 5 s · seed 42. Five-second clips, three runs, three camera setups below; the prompt is in the next reply.
1
15
Prompt, run 1: “A humanoid robot walks forward across an empty warehouse floor, then turns right and keeps walking, then turns left and keeps walking. One continuous shot, steady camera at eye level, natural daylight, real-time speed, no slow motion.”
7
My Cursor chat store measured 323.6 GB this morning, eight weeks after Cursor's GC plus my own trim and compaction took the same history down to 134 GB. I could start my agents empty tomorrow. I would lose the fastest way to reread my own work. @cursor_ai
8
1
140
@elonmusk My Cursor chat store measured 323.6 GB this morning. Most bytes are request blobs with no age-based retention. I built a fresh-start tool to park and restore the history whole; fixture-tested and dry-run, not applied live.
1
26
Replying to @Portall @cursor_ai
@XFreeze After a season of sub-agent work, my Cursor chat store measured 323.6 GB this morning. Most bytes are request blobs with no age-based retention. I built a fresh-start tool; it is fixture-tested and dry-run, not applied live.
1
20
Replying to @Portall @cursor_ai
@mattyp My Cursor chat store measured 323.6 GB this morning. Most bytes are request blobs with no age-based retention. I built a fresh-start tool to park and restore the history whole; fixture-tested and dry-run, not applied live.
19
Replying to @Portall @cursor_ai
@jediahkatz State-storage data point: my local Cursor store measured 323.6 GB this morning. Most chats were sub-agent conversations; the largest blob family is not attributed to any chat in my census. Anatomy attached.
1
38
Replying to @Portall @cursor_ai
@leerob A storage data point from my Mac: my Cursor chat store measured 323.6 GB this morning, eight weeks after a full cleanup. Most bytes are request blobs with no age-based retention. The cleanup worked; the growth outran it.
1
20
Replying to @Portall @cursor_ai
@mntruell Storage data point: my Cursor chat store measured 323.6 GB this morning, eight weeks after a full cleanup. Most bytes are request blobs with no age-based retention. The cleanup worked; the growth outran it.
21
Replying to @Portall @cursor_ai
@ericzakariasson IDE-side storage data point: state.vscdb measured 323.6 GB on my Mac this morning. Most value bytes are request blobs with no age-based retention. Anatomy attached.
10
Replying to @Portall @cursor_ai
Honest limits: by count most chats are sub-agents; by message bytes the largest chats are mine. The request blobs, the biggest family, are not attributed to any chat in my census, so I cannot say whose they are. Read-only census on Cursor 3.22.7, macOS, measured Oct 28.
1
25
@benln the data point for an automatic retention setting: my Cursor chat store measured 323.6 GB this morning. The breakdown and the fresh-start tool's test status are in this thread.
19
We posted what our Bots do before we could show what they did. Said / Done reads a draft about your Bots, splits it into claims, grades each against the files on this computer as FOUND, PARTLY, NOT FOUND, or NOT CHECKED, returns one card and one line TO MAKE IT TRUE, never rewrites the draft, and runs on Grok Bot, where Bots on one account share one computer and read the files the others left. On 2026-09-26 at 12:09 PT our two posts and one reply graded 15 of 22 hold, and at 12:22 the same texts graded 18 of 22, after two files landed on this computer, one of them declaring the roles the posts had named. At 02:04 PT on 2026-09-30 the same 18 of 22 hold, 3 partly, 1 not checked, no grade flipped. x.ai/bot/aPYsxOcUcjpTAGmiavS…
1
1
62
Fictional sample: the seed run graded a made-up draft 10 of 12 hold, 1 partly, 1 not found; TO MAKE IT TRUE: a receipt for ticket 11. First run writes audit/CONTRACT.md, audit/SAMPLE/, and audit/card.md.
14
Sample is fictional (three made-up sellers, no live store). Short hash d102bd04. CHECK: match d102bd04 2026-09-26 01:02 PT. Empty owner-decision row is the point.
10
Four Grok Bots ran. Three receipts came back. One Bot never got the handoff. Their manager is a ledger, not a chat. One job per Bot, every receipt graded by a separate Bot before I read it, a scoreboard, a name on who went quiet. Three files, free: x.ai/bot/0uOnsCTsoM8Rx9_yh2k…
We’re inviting one Grok Bot user to a Starship launch. You could win the invite by sharing the ways you've integrated @Bot into your work. The invite includes a plus one. How to submit and some examples we love:
3
3
193
A manager Bot that talks all day rereads its thread every turn, so cost grows as history squared; a ledger grows only with the work. Thursday in my own account: 3 jobs at 12:40, 3 receipts, 3 grades by 13:00. One Bot never got the handoff. I found it myself. Board names it now.
1
27
The manager my Bots have now has never said "on it".
11