postl.ai

an AI lab
in an attic.

postl.ai — a one-person AI research lab, self-hosted in an attic in the Piestingtal. Fully encrypted. Zero cloud.

184 cores / 368 threads44,544 GPU cores90 GB VRAM 1.3+ TB RAM120 TB storage100% Linux
▾ descend into the valley ▾

The fleet

The iron, sized by muscle.

Each tile is a real system — the bigger the tile, the bigger the iron. 184 cores / 368 threads, 1.3+ TB RAM, five NVIDIA cards totalling 44,544 GPU cores & 90 GB VRAM, 120 TB of encrypted storage. Tap a tile to open it up.

The minds

Watch them think. At their real speed.

One uncensored flagship LLM plus the full generative stack — text, image and video, the whole ComfyUI pipeline, served entirely on the machines above. No tokens, no frames leave the valley. Tap the orb — it'll say something else.

GLM-5.2
orakel · abliterated · Q8 · ~800 GB
9.1 tok/s

demo · uncensored · full Q8 precision

generated · 720p · 5s
Wan 2.2
lab · RTX 5090 · text-to-video · quality-first
720p · 5s clip

~4 min per clip · the pretty one

generated · 768×512 · 5s
LTX-Video
lab · RTX 5090 · text-to-video · near real-time
768×512 · 5s clip

~12 s per clip · the fast one

FLUX.2
lab · RTX 5090 · text-to-image
1024² ~6 s
generating step 0/28img #1

~10 images/min · the sharp one

SDXL
lab · RTX 5090 · text-to-image · the workhorse
1024² ~2 s
generating step 0/25img #1

~30 images/min · batch all day

▸ full model fleet, benchmarks & roadmap: /models

no tokens
leave the valley.

postl.ai is the front door of a fully self-hosted machine hall — in an attic in the Piestingtal. Owned hardware, encrypted end to end, no cloud in the loop.

t.postl@pm.me

ps: there's usually a spare U in the rack. the rack likes company.