Runner, a local LLM engine, handles tool calls that are truncated by token limits by completing them as valid JSON and returning a parseable result, allowing agent loops to continue instead of failing or retrying from scratch like other engines.
Otis is an open-source AI agent that runs local and hosted open-weight models with privacy-focused design. It automatically recommends and downloads appropriate models based on hardware and supports llama.cpp, Ollama, LM Studio, and Nvidia PAIR.
A Spanish-language article about setting up local AI (LLMs) with Emacs for programming projects. The author describes installing llama.cpp from source, selecting the Qwen 3.5 9B model, and troubleshooting compilation and library linking issues on OpenSuse Tumbleweed.
ThreadShelf is a local-first application for archiving and searching AI conversations from multiple providers (ChatGPT, Claude, Google AI Studio, OpenRouter, LM Studio, Grok) using semantic search and semantic embeddings stored in LanceDB. Users can export chats as JSON, search by meaning through a web UI or API, and optionally continue conversations using local llama.cpp or external OpenRouter.
A developer built Tetris using a local coding agent running open-weight models offline on a laptop with plain HTML, CSS, and JavaScript. The project succeeded technically but revealed a critical failure: the model repeatedly crashed when generating the game logic file due to context window misconfiguration (32768 tokens set but insufficient for the actual task), causing 70 minutes of retries with zero progress despite correct reasoning.
Eris is a local LLM agent that evolved through three iterations of tool selection: from grep-based text parsing, to keyword lists and embeddings, to a final approach using embeddings with a policy layer and GBNF grammar constraints. The system selects from ~50 tools for a 14B model by embedding user text, comparing against precomputed tool vectors, and feeding the filtered tool list to both the prompt and grammar to ensure consistency.
A software architect with twenty years in search and data systems discovered llama.cpp and the leaked Llama model, sparking a journey into local LLM experimentation. Building tools like BottyBot and RAGTag, they evolved from simple model testing toward agentic retrieval systems that treat databases as tools for intelligent agents rather than primary products.