AI

PageIndex searches long PDFs with no vector database

Vectorless RAG tool PageIndex builds a table-of-contents tree from a long PDF and lets the model reason its way to the right section. No chunking.

Outlines stops your AI from ever writing broken JSON

Outlines structured generation blocks invalid tokens while a model writes, so JSON matches your schema by construction instead of by retry and repair.

Best AI audio tools 2026

Best AI audio tools 2026 for transcription, voice cloning, cleanup. Where a free hosted tier wins and where running Whisper yourself wins outright.

OpenMMO lets AI agents play by the same rules as you

Open source MMO for AI agents OpenMMO puts bots and humans on one WebSocket protocol, in a 32 km procedural world with server-side dice-roll combat.

Lightpanda is a headless browser that fits in 123 MB

Headless browser for AI agents Lightpanda is written in Zig, peaks at 123 MB where Chrome needs 2 GB, and speaks CDP so Puppeteer connects unchanged.

turbovec fits a 31 GB vector index into 4 GB of RAM

Quantized vector index turbovec squeezes a 31 GB corpus into 4 GB and beats FAISS FastScan by 19 to 31 percent on ARM, with no separate training step.

wigolo gives your coding agent web search with no API key

Local web search MCP server wigolo gives coding agents search, fetch, crawl, and extract with no API keys, on-device reranking, and no cost per query.

llmfit tells you which local models your machine can run

Which LLM can I run on my hardware? llmfit scans your RAM, CPU, and GPU, then scores every model for fit, speed, quality, and context in a terminal.

Opus 5.5 is the Claude comeback Reddit was waiting for

Opus 5.5 Reddit threads call it the Claude comeback. It finally talks like a person, limits stretch further, and Fable 5.1 still wins some coding tests.

Web dev

The best object storage for SaaS in 2026

Best object storage for SaaS in 2026, priced on three workloads: R2 bills $2.85 where S3 bills $31.60, B2 is cheaper still, and Hetzner wins EU media.