Open · vision-native · local-first

Build vision-native agents.
From detection to understanding.

VAOS transforms camera events into structured reasoning, validation,
memory, and decision-making running locally on your hardware.

Quickstart→ Read Docs View Runtime
$ pip install vaos && vaos run --camera 0
Architecture

The Missing Layer

Traditional vision systems jump from detection to decision.
VAOS introduces state, transition, validation, and memory.

The Missing Layer — VAOS architecture diagram
Why VAOS

Detection answers:
"What is in this frame?"

Decision requires: "What changed?"

Decision Requires — VAOS reasoning diagram
Observe sensor
→
Structure scene
→
Think cognition
→
Validate validate
→
Act actions
→
Learn memory
# what-is-vaos

One coherent runtime, six layers

Sensor → reasoning → action — all running on your edge hardware. No cloud round-trips. No proprietary lock-in.

Documented end-to-end in the runtime flow on GitHub — Observe → Structure → Think → Validate → Act → Learn.

Full architecture→
VAOS Live Platform Signal
LIVE
⭐
—
GitHub Stars
LIVE
🧑‍💻
—
Contributors
LIVE
🚀
—
Releases
LIVE
🔀
—
Forks
# your-first-agent

Five lines.
One vision agent.

VAOS gives you a complete perception → reasoning → action loop out of the box. Bring your camera, swap the agent logic, deploy.

Read quickstart→ More examples
example.pypython
from vaos import Runtime, Agent

runtime = Runtime(camera=0, hardware="rk3588")

async def on_scene(scene):
    if scene.event.type == "restricted_zone_entry":
        await runtime.action("notify", target="safety-team")

runtime.attach(Agent(on_scene=on_scene))
runtime.start()
# ecosystem

Everything developer-facing,
in one place

📖

Docs

Concepts, API reference, deployment guides, and architecture notes.

docs.vaos.online →
⌨

GitHub

Source code, issues, releases, and PRs — the engine room of the ecosystem.

github.com/vaos-online →
💬

Discord

Join the community. Real-time access to the core team, contributors, and integrators.

discord.gg/Mg7TztrU →
📦

SDK

Python, REST, WebSocket. Plugin spec for extending the runtime.

sdk.vaos.online →
🧩

Plugins

Extend perception, reasoning, and action layers. Plugin marketplace + spec.

plugins.vaos.online →
🗺️

Roadmap

Public roadmap. Shipped, in-progress, planned. Vote on what comes next.

roadmap.vaos.online →
GitHub Org → Docs & SDKs ↗ Discord Community
# why-vaos

Why developers choose VAOS

// 01

Hardware-neutral

Same code, any board. RK3588, Jetson, Pi 5, ARM SoCs. Zero lock-in.

// 02

Sensor-aware

Reliability scoring built into perception — calibration, lighting, occlusion handled.

// 03

Local-first

No cloud round-trips. Full privacy. Works offline. Regulation-ready.

// 04

Open Scene Contract

Structured, machine-readable output. Integrate from any language.

// 05

Plugin runtime

Extend perception, reasoning, action with first-class plugins. Community-owned.

// 06

Persistent memory

Scenes, actions, outcomes — continuous local intelligence. Sites learn themselves.

# changelog

Latest releases

v2.0

Open Ecosystem Launch

Scene Contract 1.0 · SDK 1.0 · Governance 1.0 published. vaos-online GitHub organization open. Dashboard demo, replay traces, and ADRs available. Explore on GitHub →

2026-06-14

v0.9.2

Sensor Manager v2 + multi-camera fusion

Patent-pending sensor reliability scoring released as a stable API. Multi-camera fusion at 30 FPS on RK3588. Release notes →

2026-04-21

v0.9.0

Open-source release

VAOS Core MIT-licensed. Full Python SDK, REST/WebSocket APIs, and plugin spec public. Foundation for the v2.0 ecosystem. Read the announcement →

2026-01-30

Full roadmap & changelog→
# contributors

Built by the community

VAOS is governed under open-source principles, initiated by eye3.ai, and built by an active developer community.

JL
RM
SP
KH
YC
AV
NT
+42

Join the community→