Mini-AGI is a continual learning byte-level language model that assembles its own architecture, trains from scratch on a single 8 GB VRAM GPU and keeps learning from everything it reads. Weights live as ordinary files on disk and page onto the card as needed, so parameter count is bounded by free disk space rather than VRAM. At one snapshot it held 1.26 billion parameters while activating about 345 million per byte, and the author calls it a small toy-level model rather than a frontier one.
Be the first to rate Mini-AGI. Your feedback helps other users decide.
Fit a growing model on a single 8 GB GPU with disk-backed experts
Keep teaching the model from your own notes and files as you read
Clone the training and serving scripts and replicate the published runs
Similar tools in the same category — compare side-by-side
Janus is a Go binary that runs GGUF models through the Vulkan API, so the same local model runs on A...
Engrim is a local-first memory engine built on SQLite that gives AI command-line tools a shared, per...
Offrun is a workspace for managing coding agents, letting you keep several of them under one interfa...
Guides and comparisons about code tools
A roundup of 18 AI tools launched between August and October 2026 that reached Hacker News front page: coding agents, guardrails, on-device models, dictation.
How development environments built around delegation differ from autocomplete-era assistants, and what Windsurf and Google Antigravity each do well.
What today's AI app builders can genuinely deliver — working full-stack applications from a description — and where they still hand the problem back to you.
How AI review tools differ from linters and from human reviewers, what they reliably catch, and how to add one without annoying your team.
Cursor vs Zencoder vs Greptile compared for 2026: autocomplete, repo-level agents and AI code review. Which AI coding tool fits how your team actually works?
