SimplexAI's Linger Runs a Full AI Agent on Your PC for Free
Linger, a new open-source personal agent, runs on consumer PCs. It handles memory, agent teams and email triage locally using a MoE engine that fits a 125B model onto a 12 GB GPU.
- Linger is an open-source personal agent that runs agents, memory and models locally, with no token fees.
- Its PolyStrata engine keeps busy MoE experts on GPU and the rest in RAM or SSD.
- Reported speedups over llama.cpp: 2.7 to 3.5x for prompt reading, 1.5 to 1.8x for writing.
- A self-updating Markdown wiki memory scores 95% on LoCoMo (self-reported), just ahead of Zep and Mem0.
- It runs on 16 GB Apple silicon Macs or 8 GB NVIDIA GPUs with 32 GB RAM, and also works with Claude Code and Codex.
- Caveats: the 125B model on 12 GB is slow, some features need the cloud, and the license has hosting conditions.
SimplexAI Labs has published the source for Linger, a personal AI agent that runs its model, memory, tools, and orchestration on a user’s computer. Local model execution removes per-token inference charges, while optional cloud runtimes can handle more demanding jobs. The repository crossed 1,000 GitHub stars within roughly its first week.
The company describes Linger as an “open-source personal agent with a Claude-class brain running locally on your PC.” That description needs qualification. “Claude-class” is a vendor claim supported by the team’s own tests, and the license adds a commercial-service restriction to Apache 2.0. Connected apps, voice, and background memory processing also contact external services.
Another Me takes the keyboard
Linger’s primary agent, Another Me, works with local files, desktop applications, signed-in Chrome sessions, an inbox, and a calendar. Because it operates through the user’s existing sessions, its actions inherit the permissions attached to those accounts. According to the documentation, the agent can:
- Maintain shared memory: It records information about people, clients, projects, and topics for every connected agent to search.
- Control the computer: Agents can open applications, use the signed-in browser, access files, and run shell commands.
- Monitor connected services: It watches the inbox and calendar, then sends a message when an item needs attention.
- Delegate background work: An agent can move a long-running job into a Task while continuing the conversation.
- Coordinate agent teams: Another Me can assemble multiple agents inside a shared Room.
Linger separates its interface and coordination layer from the Runtime, the agent program that executes a request. Supported runtimes include Claude Code, Codex, and Pi. Pi is the default for local models, while existing Claude Code or Codex subscriptions can participate in the same Room. Developers can therefore route routine work to local inference and retain a cloud runtime for harder tasks.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.