An AI that acts on your OS.
Not one that talks about it.
Ask "diagnose my disk" — BATEN Shell runs the PowerShell, reads the output,
and tells you exactly what it found. Every action is Sentinel-validated
and cryptographically sealed. Your data never leaves your machine.
4
Interaction modes
0
Cloud dependencies
◆
Seal per action
$0
With your own API key
BATEN Shell [Agent] > Find the 10 processes consuming the most RAM right now
◆ Seal: d7:3a9f… · timestamp · OS state · result hashed
BATEN Shell [Agent] > Find all Rust target/ folders > 500MB and calculate total recoverable space
✅ B12 — VERT · 0% — read-only · 19s
baten_nexus\batenpsi\target 25.31 GB
baten\baten_browser\target 4.19 GB
baten_nexus\psi_server\target 3.79 GB
landing\server\target 2.32 GB
[12 more folders...]
✓ Total recoverable: 60.32 GB. All are Rust build caches — safe to delete,
rebuilt automatically on next cargo build. Want me to clean them?
◆ Seal: d7:8c2a… · causal proof generated
Four modes. One shell.
Chat
💬
Chat
Conversation with full OS context injected — active window, gravity state, process list. No generic answers.
CMD
⚡
Command
Describe what you want in plain language. BATEN Shell generates and executes the exact PowerShell command.
Plan
🔧
Workflow
Multi-step task planner. Each step validated by Sentinel before execution. Human checkpoint at any point.
Agent
🤖
Agent
Full ReAct loop: Sense → Plan → Act → Observe. The AI reads actual results and adapts. Max 5 iterations per task.
Why this is not a chatbot
Every AI assistant on the market produces language as its final output. It describes what you could do, recommends steps, writes code you have to paste elsewhere.
BATEN Shell produces actions. It executes the command, reads the real stdout/stderr, and formulates its answer based on what actually happened — not what it predicted might happen.
The architecture is borrowed from robotics: a closed Sense → Plan → Act → Observe loop where the OS is the world model. Every iteration is bounded, auditable, and reversible when possible.
Sentinel — Built-in command guard
Before any command reaches the OS, it passes through the Sentinel layer. Commands are classified by risk level (Safe / Moderate / Dangerous) and reversibility. Dangerous or irreversible operations require explicit user confirmation — they cannot be silently executed by the agent.
Path guards, command blacklists, and custom Sentinel rules (Pro tier) ensure the AI never operates outside defined boundaries.
Causal Seal — Cryptographic proof per action
Every action in BATEN Shell generates a cryptographic hash of the full action context: the prompt, OS state before execution, the command, and the result. This chain produces a proof of causality — why the AI decided to act, what it executed, and what resulted.
No action is a black box. Every step is traceable, reproducible, and auditable.
Comparison
BATEN Shell
ChatGPT
PowerShell
Executes OS commands
Yes
No
Yes
Reads real output
Yes
No
Manual
Local / offline
Yes
No
Yes
Audit trail per action
Causal Seal
No
No
Safety guard
Sentinel
No
No
BYOK (your own API key)
Yes — free
No
N/A
Annual cost
$49 once
$240/yr
Free
Pricing
⚡ Limited — 200 seats
Founder Edition
$49 once
Early bird · lifetime licence
All 4 modes (Chat / CMD / Plan / Agent)
BYOK — use your own API key free
Sentinel command guard
Causal Seal audit trail
Ollama local AI support
Lifetime updates
⚡ 200 spots only
Standard
Pro License
$79 once
All 4 modes
BYOK included
Sentinel + custom rules
Causal Seal audit trail
Ollama local AI support
Lifetime updates
Managed cloud
Cloud Add-on
$12/mo
Add-on to any licence
BATEN-hosted AI models
No API key needed
Priority inference
Requires Shell licence
Cancel anytime
"ChatGPT Plus costs $240/year. BATEN Shell Founder Edition costs $49 once."
Paid back in 3 months. Your data stays on your machine. Forever.
Get BATEN Shell
Your AI terminal — runs entirely locally. No cloud, no subscription required to start.