Linger
Simplex AI, founded by former DeepSeek-VL lead Zhang Bo, open-sourced Linger, a local personal agent that runs a 125B model on a 12GB GPU at 45.5 tokens per second with no token fees.
Linger reads local email, calendars and files and operates the user's logged-in browser, with confirmation cards for write actions. Its PolyStrata engine spreads mixture-of-experts weights across GPU memory, system RAM and SSD, so a 125B Qwen3.8-Flash-Next model can run on a 12GB card with 64GB of RAM. The license is Apache 2.0 with added commercial restrictions, and a hosted version is waitlist only.
- Date
- Thursday 19 November 2026
- Lab
- Simplex AI
- Kind
- product
- Access
- open weights (restricted license)
- Price
- Free for local use, hosted version waitlist with no pricing disclosed
Figures
| Measure | Value | Measured by |
|---|---|---|
| Decode speed, Qwen3.8-Flash-Next 125B, 12GB GPU with 64GB RAM | 45.5 tokens/second steady-state rate, excludes first-token latency; not independently reproduced | company |
| Decode speed on 32GB RTX 5090 | 179 tokens/second vendor measurement reported by Pandaily | company |
| GitHub stars | about 1.8k as of November 20 | BigGo Finance |
Source code is open under a modified Apache 2.0 license, not model weights. Packaged installers are marked coming soon, and there is no sandbox isolation.
Sources
- pandaily.com/simplex-ai-linger-local-personal-agent-polystrata
- kucoin.com/news/flash/former-deepseek-member-zhang-bo-open-sources-personal-agent-linger-w
- finance.biggo.com/news/d1a716a0-2e53-4744-b668-bd37bce7d06a
This record was checked and corrected against its sources on 20 November 2026. How we check