colibri
A hierarchical inference engine for frontier MoE models: hot experts on the GPU (CUDA, Vulkan), warm ones in RAM, cold ones streamed from NVMe. Written in C, runs on the hardware you already own.
- 53k
- C
- Apache-2.0
- 55 min ago
JustVugg
Founder of Colibrì a tiny engine, immense model for affordable hardware.
Built from JustVugg's public GitHub profile. Not on Take My Repo yet, so there are no posts or updates here.
A hierarchical inference engine for frontier MoE models: hot experts on the GPU (CUDA, Vulkan), warm ones in RAM, cold ones streamed from NVMe. Written in C, runs on the hardware you already own.
Run large AI models across your computers and servers. Powered by Colibri. Pure C.
GPT-2-style LLM built from scratch in C/CUDA with hand-written backprop, BPE tokenizer, FlashAttention, pretraining, and SFT.
Offline-first synchronization engine for SQLite and PostgreSQL, built in Rust with column-level CRDTs, authoritative partial replicas, and secure multi- tenancy.
Open-source Legal AI workspace for evidence-grounded legal drafting, matter analysis and verifiable answers.
Memory as state, not search — deterministic, dependency-free memory for AI agents. The memory lives in a Markdown file.
From any API surface to verified AI agent tools: Go CLI + MCP + skills + SQLite + policy + proofs.
Turn documents into high-quality instruction datasets with grounding, quality filtering, deduplication, and provenance.
Wiredigg is a comprehensive network analysis tool with advanced security features designed for network administrators, security professionals, and IT enthusiasts. It provides real-time packet capture, protocol analysis, anomaly detection, and threat identification capabilities in a modern, user-friendly interface.
Any website, within reach of any agent. Captures a site's native WebMCP tools and generates the bridge from the live DOM when there are none, exposing both over MCP to Claude Code, Codex and any MCP client.
Open-source middleware for AI agents — catches mid-run failures,compresses stale context, and grows a reasoning library across runs. Any model, any framework.
An MCP server for dataset generation, fine-tuning, RL, and evaluation of LLMs — directly from your coding agent.
A lightweight and fast C++ library for building MQTT clients and brokers, with support for QoS, authentication, security, persistence, and user/session management.
Help AI agents use fewer tokens and keep more useful context. Reversible code, JSON and text compression, persistent memory, focused reads and version-aware edits.
Lightweight, agent-optimized database CLI with one-shot schema introspection, column profiling, and ERD generation.
Sandboxed code execution daemon with enforced memory, CPU, and output limits. JSON API for humans and AI agents.
FlappyBri: colibri's hummingbird game, played live by a decision model through POST /v1/systemone
Hosted Kubernetes control planes that sleep: one Go binary runs many upstream control planes and keeps only the active ones in memory.
DuffyDiff is a modern, user-friendly text comparison tool built with Python and Tkinter. It allows users to easily compare two text files, navigate between differences, and merge content between files.