Last updated: 11/22/2026Last verified: 2026-11-19
Open-source personal agent workspace that runs on your own computer: a personal agent called Another Me keeps a searchable memory of your people and projects, builds teams of agents in Rooms, and drives your desktop apps and signed-in browser while local Qwen, DeepSeek or Gemma models run on your GPU. Free to self-host under a modified Apache 2.0 licence.
What is Linger?
Linger is a personal agent workspace from SimplexAI that you host on your own computer. Its personal agent, Another Me, keeps searchable memory about people, clients and projects, and Rooms let an agent plan a team from team.md so several agents can work on one job. Agents run through a Runtime you install yourself - Claude Code, Codex or Pi - and act with that Runtime's access on your files, your signed-in apps and your shell. Local models run through llama.cpp or the vendor's PolyStrata engine, with Qwen, DeepSeek and Gemma listed on the vendor's site. The desktop app covers Apple silicon Macs, Windows 11 x64 and Linux x64, and Linger can also be installed with Docker Compose. On the official site the three download buttons still read "Coming soon"; the installable packages are published on GitHub Releases. Linger is free to self-host under a modified Apache 2.0 licence, and hosted access is waitlist-only with no published price. These details were verified from the official site, documentation, repository and licence; the app was not installed or tested.
Key Features
- Shared memory: Another Me keeps information about people, clients and projects that the workspace's agents can look up.
- Rooms and team.md: an agent plans a team in team.md, then several agents work on the same job, with approvals over what they do.
- Local inference: llama.cpp or the vendor's PolyStrata engine, with Qwen, DeepSeek and Gemma models listed by the vendor.
- Runtime-based agents: Linger starts the agent CLI you already use - Claude Code, Codex or Pi - under your own account.
- Computer access: agents use your files, your signed-in apps and your shell; the docs state the limits are server checks and approvals, not a sandbox.
- Confirmations for writes: connected apps such as Gmail, Slack and GitHub are written to only after your confirmation, according to the docs.
- Voice options: a server can enable meeting recording, dictation and Grok Live voice mode using your own provider keys.
- Deployment: desktop app for Apple silicon Mac, Windows 11 x64 and Linux x64 (glibc 2.35+), or a Docker Compose server built from source.
Best For
Pricing
The official site has no pricing page (/pricing returns 404) and its macOS, Windows and Linux download buttons read "Coming soon". Linger is free to self-host under a modified Apache 2.0 licence; its two additional conditions require written authorization to offer Linger to third parties as a hosted or managed service, and forbid removing or changing the LOGO or copyright notice in the frontend. Hosted Linger is waitlist-only ("Early access to hosted Linger", invites sent in waves, free credits if picked) with no price published. Local inference is marketed as "$0 token fees", and the vendor's own figures - a 125B model on a 12 GB GPU and about 5x faster than a standard local setup - are vendor benchmarks, not measured by this listing. Hardware, electricity and any Runtime subscription are separate costs.
Pros & Cons
Pros
- One workspace for personal memory, multi-agent Rooms and computer control.
- Agents work through the CLI you already subscribe to, so provider limits and charges stay with that subscription.
- Supported local models run on your own hardware, and setup checks available memory before proposing a plan.
- Documented first-launch steps for macOS and Windows cover the unsigned-build prompts.
- The licence conditions and install requirements are published in the repository and docs.
Cons
- Site download buttons still say "Coming soon"; packages come from GitHub Releases, and the docs state they are not yet signed with a publisher certificate macOS and Windows recognize, so both block the first launch.
- Intel Macs are not supported, and an 8 GiB Mac gets no automatic local-model plan; local models need Apple silicon with 16 GiB or more unified memory, or an NVIDIA GPU with 8 GiB or more VRAM.
- Agents act with your own authority on your files, signed-in apps and shell; the docs say none of the limits is a sandbox.
- Hosted access is waitlist-only and has no public pricing.
- Vendor speed and model-size figures are vendor benchmarks; this listing did not measure speed or task quality.
Alternatives
If you only need a private chat window over local models, a narrow chat UI avoids the Runtime, Rooms and hardware-planning setup Linger adds.
If one agent CLI already covers your work, Linger's extra value is shared memory, Rooms and computer control rather than the agent itself.
If you do not want to install and maintain software on your own machine, a hosted workspace removes local hardware requirements but keeps your data off your computer.
FAQ
Details
Platform
Features
- A personal agent called Another Me that keeps a memory of your people, clients and projects, searchable by every agent you run
- Rooms where an agent assembles a team of agents from a team.md plan and each of them works the same job on your machine
- Local inference through llama.cpp or the vendor’s PolyStrata engine, with the vendor listing Qwen, DeepSeek and Gemma models and picking a plan for your hardware
- Agents can open desktop apps, work in your signed-in Chrome and run shell commands, with confirmation cards held for writes other people will see
- Meeting recording with a live transcript, dictation, and hands-free voice conversation
- Ships as a desktop app for macOS (Apple silicon), Windows 11 x64 and Linux x64, as a Docker Compose server, or run from source
Languages
Known limitations
- The official site’s “Download Linger.” block still labels macOS, Windows and Linux as “Coming soon”; the actual installers only exist on GitHub Releases (desktop-v0.1.4 on 2026-11-19) and the docs state they are not signed with a publisher certificate, so macOS and Windows block the first launch
- Intel Macs are not supported, and a local model needs either an Apple silicon Mac with at least 16 GiB of unified memory or an NVIDIA GPU with at least 8 GiB of VRAM — an 8 GiB Mac gets no automatic local-model plan
- It does not include an agent runtime: Claude Code, Codex or Pi must be installed and signed in, and the docs state that Pi runs without the Guardrail review
- The docs say plainly that none of this is a sandbox — an agent acts with the runtime’s access to your files, signed-in apps and shell
- Hosted Linger is waitlist-only with no published price, so the cloud path cannot be compared yet; the self-hosted app is free but speed depends on your own hardware
- The speed numbers (a 125B model on a 12 GB card, up to ~5x a standard local setup) are the vendor’s own PolyStrata benchmarks and were not tested for this listing








