simplexai-labs/linger

What it is

Linger is a personal AI agent that runs on your own machine instead of a cloud account. It is aimed at people who want an assistant with access to their files and apps but do not want their memory or context living on a provider's server. The agent can search Files, Mail, Calendar, Slack and Docs, keep long-term memory in wiki pages that every agent can query, and work proactively on inbox, calendar and Slack follow-ups.

The local inference story is built on PolyStrata, a MoE inference layer layered over the Strata engine (PolyStrata, Strata). The README claims a 125B-parameter model running on a 12 GB GPU by keeping hot experts in VRAM, warm experts in system RAM and cold experts on SSD, with 45.5 tokens/s reported on an RTX 3060 12 GB / 64 GB RAM. Treat those numbers as the project's own claims rather than an independent benchmark.

Beyond single-agent use, Linger supports agent teams: a job is written into a shared team.md that a research agent, a copy agent and a visuals agent all follow.

Quick start

The desktop app is the shortest path for one person on one computer. Download the package for your platform from the releases page (GitHub Releases):

• macOS Apple silicon: Linger-<version>-arm64.dmg
• Windows x64: Linger-<version>-windows-x64-setup.exe, or the -windows-x64-portable.zip if you do not want to install
• Linux x64: Linger-<version>-linux-x64.tar.gz (requires glibc 2.35+)

Verify each file against SHA256SUMS.txt in the same release, then install and open the app. On the setup screen, Quick Install provisions a Linger server on the same machine and connects the computer to it.

For a shared server or a source build, use Docker Compose instead. You need Docker with Compose and at least 5 GiB of memory for the VM:

git clone https://github.com/simplexai-labs/linger.git
cd linger
cp .env.example .env
docker compose up --build --detach --wait

Create the owner account, replacing the test credentials with your own:

printf '%s\n' '<password>' | docker compose exec -T teamhub \
  teamhub-account create-owner --email <email> --password-stdin

Open http://localhost:3000, sign in, and follow the connection commands the app shows. If the server will serve other machines on your network, the Docker Compose install guide covers networking and enabling voice.

Using it

Several connected apps exchange data with their own services even in local mode, so "local" here means the model and memory stay on your device, not that every integration is offline. In local mode, personal data and memories stay on the device; data pulled from connected apps is still handled by those apps' services.

There is no Intel Mac package; that hardware needs the Docker Compose route. The owner account is created with a test password in the README example, and the docs say to change it under Settings › Security before exposing the server.

The runtime is configurable: the runtime guide covers Claude Code, Codex and Pi, running on either local models or a subscription. Self-hosting and troubleshooting are documented in the repo's self-hosting notes. Official documentation lives at docs.heylinger.com.

The repository is under the Linger Open Source License, based on Apache 2.0 with additional conditions (LICENSE); third-party code and unlicensed logos/images are listed in NOTICE.

HotRepo's take

Linger is a very young project: created 2026-11-10, it entered HotRepo's new-project board on 2026-11-12 with 12 stars gained over a roughly six-hour window, reaching 112 stars and 17 forks. The best all-language daily rank is not recorded, so there is no evidence yet of broad trending traction. The interesting tension is between the ambitious local-agent framing and the delivery model: the primary desktop path hands users a prebuilt installer or portable archive, and the local-inference performance figures come from the project itself. The self-hosted Docker path and the third-party NOTICE give readers something concrete to inspect before trusting the numbers.