Žinutės

Claude Programavimas / agentai

windows nebent global memory

🙂

va taip atrodo pas mane 😄 ( neina share is codex ) visada dirbau su claude bet

Programavimas / agentai

https://gist.github.com/blinkinglight/3c8dd850223a4f50a62083715aa73556

recall.sh

GitHub Gist: instantly share code, notes, and snippets.

GistPerskaityta 2026-09-09 15:03

Programavimas / agentai

atmintis panasu kad tinka ir codex'ui

vien zo, reik kazkaip issprest windows vartotoju LLM memory problemas, gyvenimas pageres 100x 😄

❤1

SlaptasBotas

WSL2 naudot :)

Darius@Kiaulakis.lt #AiOpsCatalyst

:))

sudetinga

Už ką dabar sumoka ir įdarbinant tikisi iš žmogaus..

Soft Skills for AI Work: What the Labour Market Pays For
https://www.linkedin.com/pulse/soft-skills-ai-work-what-labour-market-pays-stanislav-shulga-scpdf?utm_source=share&utm_medium=member_android&utm_campaign=share_via

Soft Skills for AI Work: What the Labour Market Pays For

I took up personal time management and planning in the mid-2000s, while working at a project management company. A professional deformation of sorts: if you want to manage something, you might as well start with yourself.

Perskaityta 2026-09-09 15:03

vien zo, palinkesiu kantrybes 😄 panasu kad galim issprest AI memory problema 😉 isgydoma ta skleroze 🙂

❤1

Claude

idomumo delei, galit gal kas imest savo statsus claude?

pvz pas mane 2.2billion tokenu yra "cache read"

Programavimas / agentai

neturiu pasirinkimo, tai bandau codex 5.6-sol su xhigh 😉

mh. atrodo jog po "context compaction" sekmingai viska atsimine ir pratese darba

idomumo delei, galit gal kas imest savo statsus claude?

PauliusAutomatizacija / integracijos

ojo 😄

pas tave toks mazas input ir toks didelis output santykis. kuo pagal si santyki skiriasi mano ir tavo darbas?

as per daug sedziu ir menedzinu or pas tave automatiskai jisai viska daro?

mazos iteracijos

input kaip suprantu promptai

tai jie trumpi

ir skill pora kurie apraso visa tech stack tai initial one shot viska o poto agile 😄

o daugiau viskas is memory

su kiekvienu projektu vis daugiau memory kaip daryt ir kas veikia / neveikia

tai kaskart oneshot'ina - vis geriau 🙂

as tiesiog savo esmini skill pats rankytem susirasiau 😉

ir skill pora kurie apraso visa tech stack tai initial one shot viska

supratau. it tikrai, as beveik neoneshotinu nieko. sedziu ir laika savo naudoju.

PauliusAutomatizacija / integracijos Programavimas / agentai

na, dar cia del to, jog as mazai ka su GUI darau 🙂

visas darbas algoritmai ir backend 😉

pagrinde*

siaip dariau testa

su memory subagenta spawnino tai is "LLM ismatymo" vis dar darytu tas pacias klaidas ne pagal paskutine bibliotekos versija 😄

na palyginimui taskas. kaip be memory ir kaip su memory koda generuoja

na palyginimui taskas. kaip be memory ir kaip su memory koda generuoja

ar galetum pasidalinti izvalgom:

Ask him:

- Memory mechanics - how agentsmemory stores and retrieves (schema, wings, dedup), and are entries written automatically or hand-curated?
- The recall check - exact prompt/procedure: what does it compare, what happens on a miss?
- Verification - how does he validate a one-shot without sitting in the loop; what's his rework rate when a one-shot misses?
- Staleness - how does memory stay current when a library version changes; does anything expire or get overwritten?
- Limits - would his approach survive GUI-heavy work, or is it backend-only in practice?

👍1

na, cia visa sistema 🙂

Mindaugas ŽvirblisProgramavimas / agentai

cia bootstrap failas, v1 draft : https://github.com/atvirokodosprendimai/agentsmemory/blob/main/setup.md

👍2

cia bootstrap failas, v1 draft : https://github.com/atvirokodosprendim

puiku, aciu

nezinau ar pades tik

nes reik GPU ir TEI (embeding/ rerank) 😄

Combined mean for these stress cases:

Ollama semantic: 26.172s
TEI semantic: 4.287s
TEI lexical: 1.181s
TEI made semantic 6.11x faster than Ollama.
Semantic still cost 3.63x lexical with TEI.

ollama *** and must die 😄

liudesiukas ollama tiesa sakant 🙂

1 procesas utilizuoja tik ~20% gpu, kai text embeding interface - 1 procesas uzmusa gpu 100% 😄

tai kad isnaudot gpu su ollama reik 4-8 procesu ir dar balansavimo prieki 🙂