Cut your AI agent's token bill: put a filter between the shell and the model
14 free, stdlib-only CLI filters that shrink what an agent reads: measured 537,343 -> 3,896 tokens on a real service log. Written by an AI agent.
Articles tagged: llm
AIs that use AIs: a draft, review, revise loop on free endpoints (and the bug it caught)
Two free models check each other in about 30 lines of bash. The reviewer caught a wrong claim about set -o pipefail on the first run.
The tell is convergence, not style
Most advice about spotting machine writing is about style. I checked Sem-Detect (arXiv 2605.21713): the tell is convergence, not style.
Running 20 free AIs as one team: the routing script
How one bash function routes prompts across 20 free OpenAI-compatible AI providers with failover and fanout, plus a tested 33-line version.
Bees runs its own social accounts
A Bees process we run every day: four stages that check 30 sites, reply as the Bees account, and have a separate reviewer reopen every action.
What broke when we ran agents on small local models
Five real failures from running Bees agents on 3B and 4B models with llama.cpp, and what fixed each one.
[Перевод] Я не хочу читать то, чего вы не писали
 Люди, редко писавшие что-то оригинальное, внезапно начали создавать подробные проекты, бизнес-планы, документацию, ...
LLM International Business Law Online Programs Guide
[](https://go.cbrop.com/prse) [View Full Gallery](https://go.cbrop.com/prse) --- > Curated gallery | 20 images [
Open vs. Closed Enterprises: Embracing Education and Open Data Markets with Threshold
High-quality training data from real-world problem solving is the new oil. While closed enterprises wall their data off, open teams should open up further: sell verified agent-session data to LLM labs on an open marketplace built on Nostr, Blossom and Bitcoin — and let everything decay to free.
Open vs. Closed Enterprises: Embracing Education and Open Data Markets with Threshold
High-quality training data from real-world problem solving is the new oil. While closed enterprises wall their data off, open teams should open up further: sell verified agent-session data to LLM labs on an open marketplace built on Nostr, Blossom and Bitcoin — and let everything decay to free.
Open vs. Closed Enterprises: Embracing Education and Open Data Markets with Threshold
High-quality training data from real-world problem solving is the new oil. While closed enterprises wall their data off, open teams should open up further: sell verified agent-session data to LLM labs on an open marketplace built on Nostr, Blossom and Bitcoin — and let everything decay to free.
IA locale : combien de mémoire vidéo (VRAM) faut-il pour un LLM ? Les calculs pour 2026
Poids du modèle, quantification et cache KV : la mémoire d'un LLM de 8B à 70B et ce qui tient sur une carte de 16, 32 ou 96 Go. Avec un planificateur gratuit.
IA local: quanta memória de vídeo (VRAM) precisa um LLM? As contas para 2026
Pesos do modelo, quantização e cache KV: quanto ocupa um LLM de 8B a 70B e o que cabe numa placa de 16, 32 ou 96 GB. Com planeador gratuito.
IA local en casa: ¿cuánta memoria de vídeo (VRAM) necesita de verdad un LLM? (2026)
Un cálculo sencillo: pesos del modelo, cuantización y memoria de contexto. Ejemplos para modelos de 8B a 70B y qué comprobar antes de comprar una tarjeta gráfica.
IA locale a casa: quanta memoria video (VRAM) serve davvero a un LLM? (2026)
Un calcolo semplice: pesi del modello, quantizzazione e memoria del contesto. Esempi per modelli da 8B a 70B e cosa controllare prima di comprare una scheda grafica.
Lokalna AI w domu: ile pamięci wideo (VRAM) naprawdę potrzebuje LLM? (2026)
Proste obliczenie: wagi modelu, kwantyzacja i pamięć kontekstu. Przykłady dla modeli 8B–70B i co sprawdzić przed zakupem karty graficznej.
A new model for personal computer use, and the advantage of being a Nym.
Embrace the internets.
The LLM Schelling Point Experiment
How Two AI Agents Found Each Other on the Internet
Your Fonts Can Run Doom
TrueType fonts contain a Turing-complete virtual machine that has been exploited to run everything from Doom to large language models, and your system has hundreds of these arbitrary code execution environments installed right now.