-
Guardrails with a small model: filtering output on your own machine
A 12-weight classifier plus 4 rules blocked 2 injection attempts and 2 secret leaks, and let 2 legitimate questions through. Runs in pure Python, no large model, no cloud. Granite Guardian 2B is Apache 2.0; llm-guard is MIT.
-
Cost per consumer: how much each service spends on models
Record each call with consumer, model and day. A script groups them and shows who spends. Most of the cost usually sits in a single consumer.
-
Ready made MCP servers to plug into your agent
The two official servers worth having run through npx with nothing installed system wide: the file server exposes 14 tools and the memory server exposes 9. Both verified over plain JSON-RPC.
-
Running a language model on CPU with llama.cpp
llama.cpp runs a whole language model on a CPU with no GPU: a 17 MB binary, a 468 MiB quantized model, and about 50 tokens per second of generation. The script downloads, measures and tests it.
-
Semantic search over your own files with sqlite-vec
A similarity query over 5000 documents answered in 1.7 ms inside SQLite, with the 160 KB sqlite-vec extension. No server, MIT or Apache 2.0 license.
You reached the end. Every post is here.