Žinutės

Issues · MemPalace/mempalace

The best-benchmarked open-source AI memory system. And it's free. - Issues · MemPalace/mempalace

GitHubPerskaityta 2026-09-09 15:06

TildeAI/TildeOpen-30b · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Perskaityta 2026-09-09 15:06

A Netflix engineer just open-sourced a tool that can cut LLM token usage by up to 95% - without changing your code 😳 Headroom, built by Netflix engineer Tejas Chopra, sits in front of tools like… | Linas Beliūnas | 237 comments

A Netflix engineer just open-sourced a tool that can cut LLM token usage by up to 95% - without changing your code 😳 Headroom, built by Netflix engineer Tejas Chopra, sits in front of tools like Claude, Cursor, Codex, and other agents as a local proxy. Before your payload hits the model, Headroom compresses the context. Not by blindly chopping it down. By using specialized compressors for different payloads: → SmartCrusher for JSON → AST-based compression for code → Tool-output and log compression → Local reversible storage of originals → Agent wrappers that make it usable without rewriting your app The headline claim is 60–95% fewer input tokens while preserving answer quality. The repo has already crossed 42K+ GitHub stars, which says something obvious: Developers are not just worried about AI getting smarter. They’re worried about AI getting expensive. Of course, compression is not free magic. Complex reasoning tasks may punish missing context. Agent loops may behave differently. Pr

LinkedInPerskaityta 2026-09-09 15:06