Agent Action Runtime · House #22,838

What it is

Calls itself “execution layer”.

• Calls itself: execution layer

“…open-source execution layer for agents”

Where it puts itself

Rules out “sandbox or credential vault”.

• Says it will not: sandbox or credential vault

“A tool returning 200 OK does not mean the real-world action succeeded.”

“The runtime does not provide a sandbox or credential vault.”

Who it sells to

In Python.

• Languages: Python

“Requires Python 3.10+.”

How it runs

Runs locally and works offline. No API. Open source. AI is the core of the product.

• Runs where: on the device, offline
• API: no
• Open source: yes
• AI role: core to the product

“Infrastructure Local-first, stateless, no server/database required”

“The project owns the action execution boundary: policy, approval, execution, retry, verification, recovery, and receipts.”

What it works with

Integrates with OpenAI. Names Codespaces, GitHub Advanced Security and 3 more on its pages.

• Integrates with: OpenAI
• Names: Codespaces · GitHub Advanced Security · GitHub Copilot · +2

“Agent Action Runtime is an open-source execution layer for agents that take real-world actions.”

“Codespaces Instant dev environments”

What it offers as proof

Live, no waitlist. Publishes docs.

“README.md -> https://github.com/MrRex168/agent-action-runtime/blob/main/README.md”