simplexai-labs/linger
LINGER: LOCAL PERSONAL AGENT 125B
Open-source personal agent running a 125B MoE model on 12 GB GPU. Remembers context, works proactively, and keeps data locally.
// KEY FEATURES
- 125B Qwen3.8-Flash-Next on 12 GB GPU via PolyStrata engine
- Auto-select best model from 49 variants for your hardware
- Another Me memory: wiki from mail, chats, and meetings
- Proactive mode with nightly recursive self-optimization
// DEEP DIVE: HOW IT WORKS
Linger operates as a local personal agent powered by the MoE engine PolyStrata: the busiest experts stay in VRAM, the rest in RAM or SSD. This enables running the 125B Qwen3.8-Flash-Next on a 12 GB GPU with 64 GB system RAM. The Model Autopilot installer scans hardware, filters 49 tested models (various quantizations and contexts) to those fitting a 64K context, then picks the smartest, then the fastest.
Key differentiator is the Another Me memory system: the agent nightly reads inbox, calendar, and Slack, structuring them into wiki pages shared across all sub-agents. Proactive mode lets the agent initiate actions: track tasks, remind, push work forward. Recursive self-optimization each night rewrites decision rules based on feedback.
Requirements: consumer GPUs (RTX 8–32 GB VRAM) or Macs (16–24 GB unified memory) plus 32–64 GB RAM. Tested on NVIDIA and Apple Silicon. No cloud keys or tokens needed — inference runs fully locally.
// USE CASES
Best for: Users needing a local AI agent with memory and proactivity
- Run 125B model locally on consumer GPU without cloud costs
- Automate mail and calendar processing with proactive reminders
- Build a personal knowledge base accessible to all local agents
New repositories every 30 minutes
REDDYX AI scans GitHub 24/7 and ships the best AI/ML/Web3 projects to Telegram.
Join on Telegram