Linger
A personal AI agent that runs entirely on your own computer — a 125B model on a 12 GB GPU, no cloud bill, no token fees.
Project intro
Linger is an open-source personal agent that runs entirely on your own computer. Instead of sending your mail, chats and meetings to a cloud service, it keeps them on your machine: a local model reads them, files them into wiki pages every night, and a team of agents works from that shared memory. You talk to one agent, and every agent knows what matters.
The headline trick is the size of brain it fits on ordinary hardware. Its PolyStrata inference engine spreads a 125B model across GPU, RAM and SSD, keeping only the busiest experts on the card, so the Qwen3.8-Flash-Next model runs at 45.5 tokens per second on a 12 GB RTX 3060. There is no cloud bill and no token fee for local inference, and a setup step picks the smartest model your machine can actually run, then counts it as installed only after it answers a real question.
It works with the tools you already use: agents open the apps on your Mac, Windows PC or Linux desktop and work in your everyday Chrome with the sign-ins you already have, and you can bring Claude Code, Codex or Pi as the runtime, on local models or a subscription. On the LoCoMo long-conversation memory benchmark the project reports 95% accuracy. It has 718 stars on GitHub with 46 forks, was created on 10 November 2026, and was still being pushed on 19 November 2026.
Two honest limits: this is young software, nine days old at this check, so expect rough edges and a moving setup. And the big local model still wants a real GPU and plenty of RAM: the 125B setup quotes 64 GB of RAM beside the 12 GB card, and smaller machines get smaller models.
Actions
Tech stack
Project info
- Project category
- AI Agents
- Project status
- Online
- Collaboration intent
- None