cd /news/artificial-intelligence/synapseos-where-the-kernel-thinks · home topics artificial-intelligence article
[ARTICLE · art-57521] src=github.com ↗ pub= topic=artificial-intelligence verified=true sentiment=· neutral

SynapseOS – Where the Kernel Thinks

SynapseOS, an Arch-based operating system with a local LLM integrated at the system layer, has been released in alpha. The OS runs a local LLM daemon as a system service, allowing the shell, compositor, security monitor, network filter, and a kernel module to communicate over a Unix socket without network calls. The default ISO embeds Mistral 7B Instruct, making AI available on first boot.

read6 min views1 publishedJul 13, 2026
SynapseOS – Where the Kernel Thinks
Image: source
  ███████╗██╗   ██╗███╗   ██╗ █████╗ ██████╗ ███████╗███████╗
  ██╔════╝╚██╗ ██╔╝████╗  ██║██╔══██╗██╔══██╗██╔════╝██╔════╝
  ███████╗ ╚████╔╝ ██╔██╗ ██║███████║██████╔╝███████╗█████╗
  ╚════██║  ╚██╔╝  ██║╚██╗██║██╔══██║██╔═══╝ ╚════██║██╔══╝
  ███████║   ██║   ██║ ╚████║██║  ██║██║     ███████║███████╗
  ╚══════╝   ╚═╝   ╚═╝  ╚═══╝╚═╝  ╚═╝╚═╝     ╚══════╝╚══════╝

Where the kernel thinks.

An Arch-based operating system with a local LLM wired into the system layer — not bolted on top.

synui, the wlroots compositor: waybar status bar, auto-hiding dock, CRT post-process filters.

SynapseOS runs a local LLM daemon as a system service and lets the rest of the system talk to it over a Unix socket: the shell, the compositor, the security monitor, the network filter, and a kernel module that exports syscall telemetry and AI scheduling hints through sysfs. No network calls, no API keys — the model lives on the machine.

It boots to synsh

, a shell where you can type a command or just say what you want, and into synui

, a wlroots compositor that knows the AI daemon exists.

Status: alpha.Version 0.1.x. This is a real, actively developed system — the author daily-drives it — but it is early, moves fast, and will break. Try it in a VM before you give it a disk.

Grab the latest ISO from Releases and boot it. The default ISO embeds Mistral 7B Instruct (Q4_K_M, ~4.1 GB), so the AI is live on first boot with nothing to configure.

To take it for a spin without touching hardware:

git clone https://github.com/velle999/SYNAPSE.git && cd SYNAPSE
QEMU_RAM=8G ./archiso/build_scripts/qemu-test.sh   # auto-detects the newest ISO

The script uses KVM when available, boots UEFI via OVMF (falling back to BIOS), and attaches a persistent 20 GB test disk. Give it 8 GB+ of RAM when the model is embedded. Kernel and boot output are mirrored to the serial console — View → serial0

in the QEMU window.

Each lives in its own directory with its own PKGBUILD

.

Component What it does
synapd
Local LLM inference daemon (llama.cpp). Owns the model; serves every other component over a Unix socket.
synsh
AI-native shell. Type naturally, or use it as a normal shell.
synui
Wayland compositor on wlroots 0.19 — tiling and monocle layouts, per-output workspaces, XWayland, layer-shell. See
synui/ROADMAP.md

synguard

synui

subscribes to.synnet

synapse_kmod

Supporting pieces: syn-install

, syn-firstboot

, syn-model

, waybar/

(status bar config), and archiso/

(install media).

  User
   │
   ▼
 synsh ─── natural language / commands ──┐
   │                                     │
   ▼                                     │
 synapd  (local LLM — Mistral 7B)        │
   │  inference over SYN socket protocol │
   ├──► synguard      security verdicts ─┤
   ├──► synnet        network policy     │
   └──► synapse_kmod  kernel sysfs       │
            │                            ▼
            ▼                    synui (Wayland)
     /sys/kernel/synapse/
     syscall_log, ai_hints, stats,
     status, config, version

synui

is a compositor written for this system rather than adapted to it. It draws its own display-settings panel, wallpaper picker, and dock; it renders an optional CRT-style post-process pass; and it holds a live subscription to synguard

's verdict feed and synapd

's activity.

Defaults (override in ~/.config/synui/synuirc

or /etc/synui/synuirc

):

Key Action
Super (tapped alone)
Start menu (the bar's SYNAPSE badge)
Super +C
Control panel — every shortcut, plus the settings, in one place
Super +Return
Open a terminal
Super +Space
Command bar
Super +Backspace
Ask the AI
Super +A
Neural activity overlay
Super +D
Display settings
Super +W
Wallpaper picker
Super +E
Toggle visual effects
Super +Escape
Menu
Super +Tab
Cycle layout
Super +Q / Super +Shift +Q
Close window / quit compositor
Super +J / Super +K
Focus next / previous
Super +H / Super +Shift +L
Shrink / grow master area
Super +F / Super +M / Super +N
Float / maximize / minimize
Super +Shift +F
Fullscreen (forces it — for games that only do "borderless")
Super +O / Super +Shift +O
Move window to next / previous monitor
Super +P
Power saving panel
Ctrl +Alt +Delete / Super +T
Task manager (processes, CPU/RAM/GPU)
Super +G
Game mode
Super +L
Lock screen
Print
Screenshot the monitor you're on
Shift +Print / Super +Shift +S
Screenshot an area (drag it out with slurp)
Ctrl +Print
Screenshot every monitor at once
Super +19
Switch workspace
Super +Shift +19
Move window to workspace

Screenshots land in ~/Pictures/Screenshots

and on the clipboard, so you can paste one straight into a chat without opening the file.

Everything starts on boot:

systemctl status synapd      # AI inference daemon
systemctl status synguard    # security monitor
systemctl status synnet      # network policy
lsmod | grep synapse_kmod    # kernel module
cat /sys/kernel/synapse/status

ISOs built with --no-model

are ~4 GB smaller and fetch the model on first boot via syn-firstboot

. You can also drop one in by hand:

cp your-model.gguf /var/lib/synapd/models/synapse.gguf
systemctl restart synapd

synsh

Any GGUF that llama.cpp can load will work; Mistral 7B Instruct is what the prompts are tuned against.

Prerequisites — an Arch (or Arch-based) host with archiso

, base-devel

, meson

, ninja

, wlroots0.19

, qemu

, and ovmf

. Budget ~22 GB of free disk with the embedded model, ~9 GB without.

archiso/build.sh

runs the whole pipeline: builds llama.cpp (pinned at tag b8272

, matching CI), packages every component through its PKGBUILD

into a local pacman repo, fetches the model, and invokes mkarchiso

.

sudo archiso/build.sh              # full build, GPU auto-detected
sudo archiso/build.sh --no-gpu     # CPU-only llama.cpp — use for QEMU-targeted ISOs
sudo archiso/build.sh --no-model   # slim ISO, model downloaded on first boot
sudo archiso/build.sh --no-clean   # reuse the previous llama.cpp build

Output lands in archiso/out/SynapseOS-<version>-x86_64.iso

.

Package builds run as the synbuild

user under /var/tmp

, because they must live outside /home

(mode 0700). A failed package build aborts the run immediately rather than resurfacing later as a confusing pacstrap error.

For the inner-loop, skip the ISO entirely:

bash build-all.sh   # builds every component against llama-staging/usr/

Bump ** iso_version** in

archiso/profiledef.sh

and nothing else — in particular, leave SYNAPSEOS_VERSION

in archiso/build.sh

alone, as it tracks the component series rather than the image. Then run archiso/publish-release.sh

.synsh

, synui

, synguard

, synnet

, and the kernel module all speak to synapd

over a Unix socket using a fixed 28-byte binary header ( synapd/include/synapd.h):

#pragma pack(push, 1)
typedef struct {
    uint32_t magic;        // 0x53594E41 "SYNA"
    uint8_t  version;      // SYNAPD_PROTOCOL_VER
    uint8_t  msg_type;     // QUERY, SYSCALL_EVENT, SCHED_HINT, CONTEXT_PUSH/GET,
                           // STATUS, RELOAD, SHUTDOWN; responses OR SYN_MSG_RESPONSE
    uint16_t flags;
    uint32_t payload_len;  // bytes following this header, max 1 MiB
    uint32_t request_id;   // echoed in the response
    uint32_t client_pid;   // sender PID, for privilege checks
    uint64_t timestamp_ns; // CLOCK_MONOTONIC_RAW
} syn_msg_header_t;        // 28 bytes
#pragma pack(pop)

GPLv2 — the SynapseOS Project.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @synapseos 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/synapseos-where-the-…] indexed:0 read:6min 2026-07-13 ·