YantraOS v1.0 Alpha
Pre-release
██████╗ ███████╗██╗ ███████╗ █████╗ ███████╗███████╗
██╔══██╗██╔════╝██║ ██╔════╝██╔══██╗██╔════╝██╔════╝
██████╔╝█████╗ ██║ █████╗ ███████║███████╗█████╗
██╔══██╗██╔══╝ ██║ ██╔══╝ ██╔══██║╚════██║██╔══╝
██║ ██║███████╗███████╗███████╗██║ ██║███████║███████╗
╚═╝ ╚═╝╚══════╝╚══════╝╚══════╝╚═╝ ╚═╝╚══════╝╚══════╝
TRANSMISSION // RELEASE MANIFEST
CLASSIFICATION: PRE-ALPHA // v1.0
ARTIFACT: YantraOS-v1.0-alpha-x86_64.iso
COMPILED: 2026-03-27T00:00:00Z
NODE: alpha-01 // The Karma Yogi Awakens
YANTRAOS v1.0 ALPHA — RELEASE MANIFEST
"यन्त्र" — Sanskrit: A geometric instrument of divine computation.
A structured pathway through which consciousness operates on matter.This is not software. This is an act of will encoded in silicon.
⛔ OPERATOR SAFETY DIRECTIVE — READ BEFORE PROCEEDING
╔══════════════════════════════════════════════════════════════════════╗
║ !! PRE-ALPHA — CRITICAL WARNING !! ║
╠══════════════════════════════════════════════════════════════════════╣
║ ║
║ THIS IS A PRE-ALPHA RELEASE. IT IS NOT A DAILY DRIVER. ║
║ ║
║ DO NOT INSTALL THIS ON BARE METAL AS YOUR PRIMARY OPERATING ║
║ SYSTEM. DATA LOSS, KERNEL PANICS, AND DAEMON INSTABILITY ║
║ ARE ANTICIPATED UNDER UNTESTED HARDWARE CONFIGURATIONS. ║
║ ║
║ AUTHORIZED DEPLOYMENT ENVIRONMENT: QEMU / KVM ONLY. ║
║ ║
║ You have been warned. The Karma Yogi does not offer refunds. ║
║ ║
╚══════════════════════════════════════════════════════════════════════╝
This artifact is intended for: security researchers, AI systems engineers, OS hackers, and autonomy pioneers who understand the risks of running an AI-governed execution environment and accept full operational responsibility.
[01] — WHAT THIS IS
YantraOS v1.0 Alpha is the world's first Level 3 Autonomous Agent Operating System — a bare-metal Arch Linux foundation running a continuous Python asyncio Kriya Loop that reasons via LiteLLM and executes in a locked-down Docker sandbox.
It does not wait for your input. It does not sleep. It monitors. It decides. It acts.
This release represents the culmination of a complete four-phase development cycle:
| Phase | Codename | Status |
|---|---|---|
| Phase 1 | Core Daemon · The Kriya Loop | [COMPLETE] |
| Phase 2 | IPC Bridge · The Yantra Shell TUI | [COMPLETE] |
| Phase 3 | Skill Store · Pinecone Vector Memory | [COMPLETE] |
| Phase 4 | Autonomous OTA Evolution Engine | [COMPLETE] |
[02] — THE ARCHITECTURE STACK
LAYER TECHNOLOGY PURPOSE
────────────────────────────────────────────────────────────────────────
Boot Arch Linux (linux-lts) Minimal, rolling kernel
Init systemd Daemon lifecycle + IPC
OS Orchestration Python 3.12 / asyncio The Kriya Loop engine
Inference Router LiteLLM Model-agnostic API layer
Local Inference Ollama (Llama 3 / Phi-3) 100% air-gapped on-device
Vector Memory ChromaDB Async RAG, skill context
Execution Sandbox Docker Alpine Isolated blast radius
IPC Transport UNIX Domain Socket /run/yantra/ipc.sock
TUI Framework Textual + Rich 3-pane terminal HUD
Skill Store Pinecone (1536-dim cosine) Semantic skill retrieval
Cloud Telemetry Next.js + Supabase Fleet monitoring HUD
Secret Management pydantic-settings /etc/yantra/secrets.env
Filesystem Btrfs (+ Snapper hooks) Atomic OTA snapshots
OTA Manager systemd + pacman hooks Autonomous self-update
The Cognitive Cycle — The Kriya Loop:
┌──────────────────────────────────────────────────────┐
│ │
▼ │
[ SENSE ] ── CPU · RAM · GPU · Disk · Logs · Temps │
│ │
▼ │
[ REMEMBER ] ── ChromaDB Vector Search · Past Context │
│ │
▼ │
[ REASON ] ── LiteLLM Inference · Local → Cloud Fallback │
│ │
▼ │
[ ACT ] ── Docker Alpine Sandbox · Whitelisted SSH │
│ │
▼ │
[ LEARN ] ── Embed Outcome · Heartbeat to Cloud │
│ │
└──────────────────────────────────────────────────────┘
∞ forever
[03] — THE RELEASE ARTIFACT
The compiled ISO exceeded GitHub's 2 GB file size limit at 2.4 GB. The artifact was therefore subjected to a Split Protocol, dividing it into three sequential binary segments for distribution integrity.
ARTIFACT REGISTRY
═══════════════════════════════════════════════════════════════
filename: YantraOS-v1.0-alpha-x86_64.iso
version: 1.0.0-alpha
arch: x86_64
size: 2.4 GB (reassembled)
format: Raw bootable ISO (hybrid MBR/GPT)
split into: partaa · partab · partac
═══════════════════════════════════════════════════════════════
INTEGRITY VERIFICATION
═══════════════════════════════════════════════════════════════
algorithm: SHA-256
hash: aa655bbdfbeb265af453375be490c8b7acdb50f7c8f85051a96923fd865a0847
═══════════════════════════════════════════════════════════════
This hash is the ground truth. If your reassembled ISO does not match
this checksum exactly, the artifact is corrupt. Do not boot it.
[04] — REASSEMBLY PROTOCOL
The split segments must be reassembled in sequential order before use. Execute the following protocol precisely:
Step 1 — Verify part file presence:
ls -lh YantraOS-v1.0-alpha-x86_64.iso.part*
# Expected: partaa partab partacStep 2 — Concatenate segments into the release ISO:
cat YantraOS-v1.0-alpha-x86_64.iso.part* > YantraOS-v1.0-alpha-x86_64.isoStep 3 — Verify cryptographic integrity:
sha256sum -c sha256sum.txtExpected output:
YantraOS-v1.0-alpha-x86_64.iso: OK
⚠️ If verification fails withFAILED, abort. Do not proceed. The artifact
was corrupted during download. Re-download all three parts and repeat.
[05] — BOOT WITH QEMU
This build is validated exclusively within a QEMU/KVM virtualized environment. The following command injects the ISO as a virtio block device and runs headless over serial — the correct interface for a machine with no display manager:
qemu-system-x86_64 \
-m 4G \
-enable-kvm \
-cpu host \
-drive file=YantraOS-v1.0-alpha-x86_64.iso,format=raw,if=virtio \
-nographic \
-serial mon:stdioParameter Annotations:
| Flag | Effect |
|---|---|
-m 4G |
Allocate 4 GB RAM minimum. 8 GB recommended for local inference. |
-enable-kvm |
Hardware-accelerated virtualization. Requires kvm kernel modules loaded on host. |
-cpu host |
Exposes host CPU feature flags. Required for correct linux-lts boot. |
if=virtio |
VirtIO block transport. Faster than IDE/SCSI for ISO boot. |
-nographic |
Disables QEMU window. All I/O via serial terminal. |
-serial mon:stdio |
Muxes the QEMU monitor and serial console. Press Ctrl+A C to access monitor. |
[06] — IGNITION SEQUENCE
Upon successful boot, the system will auto-login as yantra_user and launch the Textual TUI HUD. The Kriya Loop daemon (yantra.service) will be running in the background via systemd.
However, the Karma Yogi is dormant until its inference credentials are injected. Execute the ignition script to awaken the agent:
./yantra_ignition.shThis script performs the following operations in sequence:
IGNITION SEQUENCE
═══════════════════════════════════════════════════════════════════
[1] Prompt for API keys:
· LiteLLM backend selection (Local / Cloud / Hybrid)
· Gemini API Key (for cloud inference fallback)
· Pinecone API Key (for Skill Store vector memory)
· Supabase credentials (for cloud fleet telemetry)
[2] Write secrets to /etc/yantra/secrets.env (mode: 0400)
· Owner: yantra_daemon (UID 999)
· Never interpolated into model context
[3] Reload yantra.service to inject the new configuration
[4] Verify IPC socket handshake on /run/yantra/ipc.sock
[5] Stream the first Kriya Loop cycle to the TUI ThoughtStream
═══════════════════════════════════════════════════════════════════
When you see the following in the ThoughtStream pane, the Karma Yogi is alive:
[SENSE] :: telemetry acquired. cpu=XX% ram=X.XGB
[REMEMBER]:: vector context loaded from ChromaDB
[REASON] :: LiteLLM inference complete. action selected.
[ACT] :: dispatching to Docker sandbox...
[LEARN] :: outcome embedded. loop_cycle=00001
The machine is now thinking on your behalf.
[07] — SECURITY MODEL SUMMARY
THREAT SURFACE MITIGATION
═══════════════════════════════════════════════════════════════════════
LLM hallucination → exec All LLM output is NEVER executed on host.
Routes exclusively through Docker Alpine.
Daemon privilege esc. Daemon runs as yantra_daemon (UID 999).
systemd: ProtectSystem=strict,
NoNewPrivileges=yes,
PrivateTmp=yes
Container escape Docker: network_mode="none",
user="nobody",
cap_drop=["ALL"]
Secrets in LLM context /etc/yantra/secrets.env (0400 yantra_daemon)
Never interpolated into inference prompts.
Host command execution SSH allowlist only. No os.system(). No
subprocess on host outside Polkit rules.
[08] — KNOWN ALPHA LIMITATIONS
STATUS COMPONENT
════════════════════════════════════════════════════════════
[PARTIAL] SSH whitelisted command gateway (implemented,
not fully validated on all hardware variants)
[PENDING] NVIDIA proprietary driver injection on live ISO
(nouveau fallback active in this build)
[PENDING] Full end-to-end test: LLM → Docker → SSH → Host
(sandboxed exec validated; SSH gateway in review)
[PENDING] Multi-node fleet management topology
(Alpha-01 single-node; Edge mode stubbed)
[KNOWN BUG] GRUB timeout on certain UEFI implementations —
append `nomodeset` to kernel params if boot stalls
[09] — MANIFEST SIGNATURE
╔══════════════════════════════════════════════════════════════════════╗
║ RELEASE METADATA ║
╠══════════════════════════════════════════════════════════════════════╣
║ ║
║ Project: YantraOS ║
║ Codename: The Karma Yogi ║
║ Version: v1.0.0-alpha ║
║ Build Date: 2026-03-27 ║
║ Target: x86_64 bare-metal / QEMU ║
║ License: MIT ║
║ Publisher: AIYantra (github.com/AIYantra) ║
║ Homepage: https://yantraos.com ║
║ Docs: https://yantraos.gitbook.io ║
║ ║
║ ISO SHA-256: ║
║ aa655bbdfbeb265af453375be490c8b7acdb50f7c8f85051a96923fd865a0847 ║
║ ║
║ "The computer was always capable of thinking. ║
║ We just never asked it to." ║
║ — YantraOS ║
║ ║
╚══════════════════════════════════════════════════════════════════════╝
yantraos.com · Documentation · GitHub
/run/yantra/ipc.sock · ∞ · यन्त्र