» Tag
llm
472 postsWhy $/Token Pricing Hides the Real Cost of Frontier AI Models
Frontier AI pricing pages hide tokenizer differences that can inflate real costs by up to 73% on code like TypeScript, per new billing analysis.
AEGIS: An Open-Source, Self-Hosted Personal AI Orchestration System
Developer open-sources AEGIS, an MIT-licensed, self-hosted personal AI orchestration platform built on FastAPI, Postgres, and Temporal.
VetoBench Tests Whether AI Agent Memory Retains Rejected Decisions
VetoBench is an open benchmark testing whether AI agent memory systems retain and surface previously rejected technical decisions.
CommitBrief — AI code reviews, right in your terminal
A provider-agnostic, local-first CLI that reviews your staged changes, a historic range, or a whole GitHub pull request. Zero telemetry, no server. Free and open source.
commitbrief.comClaude Code Burns 33K Tokens Before It Even Reads Your Prompt
Wire-level analysis shows Claude Code sends 33K tokens before reading your prompt - 4.7x OpenCode, with 3.7x higher real-task costs.
Hermes Agent: A Self-Improving AI Framework With Persistent Memory
Nous Research's open-source Hermes Agent framework combines persistent memory, reusable skills, and a real multi-agent architecture for self-improving AI.
Benchmarking a Markdown Knowledge Graph as AI Agent Memory
IWE tested markdown knowledge graphs as AI agent memory using the LOCOMO benchmark, reaching 96% of a hand-built ceiling with a cheap curator model.
Anatomy of a Full Self-Hosted RAG Stack, End to End
A self-hosted RAG pipeline in full: Docling parsing, dual dense/sparse indexing, RRF hybrid search, knowledge graphs, token budgeting, and parallel ingestion.
Orchestrated Agents Over One Prompt: Lessons From a Branding Platform
BrandBrahma's team explains why they chose a four-orchestrator agent architecture over a single LLM call, separating verification tasks from generative ones for reliability.
MiniMax M2.7: Open-Source AI That Optimized Its Own Training
MiniMax's open-source M2.7 model actively shaped its own training via memory writes and skill-building, gaining 30% efficiency and matching GPT-5.3-Codex on SWE-bench Pro.
Building an AI that pentests your AI, with proof required
An engineer built an adversarial LLM harness that only reports proven exploits against a live agent, replacing vague 'might be vulnerable' pentest reports with deterministic, oracle-verified findings.