On November 19, Tony Zhang, founder of Simplex AI and an early core member of DeepSeek, announced on X that Linger is now open source: a personal agent that runs on your own Mac or PC. Its MoE engine PolyStrata keeps the busiest experts on the GPU, others in RAM and rare ones on SSD; Simplex lab figures show the 125B Qwen3.8-Flash-Next writing 45.5 tok/s on a 12 GB RTX 3060 with 64 GB of RAM, with no token fees for local inference.
Key Takeaways
- ✓Launch: Tony Zhang (Simplex AI founder; his X bio says he was an earliest core member of DeepSeek) announced Linger as open source on X on November 19; the post has about 760K views
- ✓Mechanism: PolyStrata tiers MoE weights by how often each expert is used (GPU, RAM, SSD) and promotes frequently picked experts to the GPU, so large models run on small cards
- ✓Numbers (Simplex lab): the 125B Qwen3.8-Flash-Next writes 45.5 tok/s on a 12 GB RTX 3060 and 179 tok/s on an RTX 5090; the 38.1 tok/s 8 GB laptop figure was a simulated budget, not a real card
- ✓Features: Model Autopilot detects hardware and installs the smartest model that fits; agents can drive local desktop apps and your signed-in Chrome and wait for approval before actions such as sending email
- ✓Platforms and license: installers for macOS (Apple silicon), Windows x64 and Linux x64; the license is a modified Apache 2.0 that requires a commercial license to offer Linger as a hosted service

Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background
Frontier models are trained in datacenters and rented back token by token. On November 19, Tony Zhang, founder of Simplex AI (his X bio says he was an earliest core member of DeepSeek), announced on X the open-source release of Linger, a personal agent that lives on your own computer, with local inference, no token fees and data that stays on the device.
How it works
The core is the MoE inference engine PolyStrata. Each token activates only a few experts, so the engine keeps the most used experts on the GPU, occasional ones in RAM and rare ones on SSD, promoting experts as they get picked. Model Autopilot reads the GPU, memory and bandwidth, picks the smartest model that fits with at least a 64K context and checks it answers a real question before counting it as installed.
Numbers that matter
All figures are Simplex lab measurements: the 125B Qwen3.8-Flash-Next (a 39 GB 2-bit file) writes 45.5 tok/s on a 12 GB RTX 3060 with 64 GB of RAM and 179 tok/s on an RTX 5090. The 38.1 tok/s 8 GB laptop figure came from a larger card held to an 8 GB budget, not a real 8 GB card. There is no independent reproduction yet.
What it means
Installers for macOS (Apple silicon), Windows x64 and Linux x64 are on GitHub Releases, with docs at docs.heylinger.com; the macOS build is not yet signed, so the first launch must be allowed manually. The license is a modified Apache 2.0: personal and in-house commercial use are fine, but offering Linger as a hosted service needs a commercial license.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.