Agent Sandboxes is live in Pydantic Logfire

No dark-mode styles — this email renders the same in a dark client.

Pydantic launches a new product: meet Agent Sandboxes.

Agent Sandboxes is live in Pydantic Logfire

Pydantic ( https://pydantic.dev?utm_source=newsletter&utm_medium=email&utm_campaign=launch-agent-sandboxes&utm_content=logo )

Hi,

Agents don't need a computer. They need a place to securely run code.

Today we're launching Agent Sandboxes in Pydantic Logfire: a sandbox for the code your agent writes. It's not a container or VM. It's an isolated interpreter written in Rust, powered by Pydantic Monty, our open-source interpreter.

Watch the Agent Sandboxes launch video: an excavator, standing in for a full virtual machine, lowered over a paper cup that stands in for the code your agent wrote. ( https://youtu.be/zXKlw6IQXiw?utm_source=newsletter&utm_medium=email&utm_campaign=launch-agent-sandboxes&utm_content=video-poster )

Why we built it

More agents now write a short script to call their tools, instead of calling them one at a time. It saves tokens and round trips: in our benchmark, code mode cut costs by 58% and wall time by 51% on long tasks. But that script needs somewhere to run, and the usual answer is a whole machine per session, booted, secured and billed while it waits.

What you get

* Only the tools you give it. No files, network or secrets unless you grant them. Your tools keep their own access, so the code never holds your credentials.
* Right-sized. No container image to build and no VM to keep warm. You pay per run, not per second of a running machine.
* A fresh session for every run. Cold starts are measured in milliseconds.
* Pause and resume. A session can stop at a tool call, for example while someone approves a refund, and pick up later right where it left off.
* Traced in Logfire. Every run shows up next to your agent, with the exact code the model wrote and the tools it called.

Where it fits

Agent Sandboxes runs Python today. If your agent needs a shell, pip install or a full filesystem, use a container or VM sandbox. For the short scripts most agents write, this is the right size.

Start free

Every Logfire plan includes 50,000 Agent Sandboxes runs a month.

Turn on Agent Sandboxes ( https://pydantic.dev/logfire/agent-sandboxes?utm_source=newsletter&utm_medium=email&utm_campaign=launch-agent-sandboxes&utm_content=turn-on-button )

Live demo · Online

See Agent Sandboxes live with Samuel Colvin

Wed, Nov 18 · 12:00pm ET / 17:00 UTC · 30 min

Samuel Colvin, CEO and founder of Pydantic, runs agent-written code in Agent Sandboxes, then follows each run through its Logfire trace. Bring your questions.

Save your spot ( https://pydantic.dev/events/agent-sandboxes-live-demo?utm_source=newsletter&utm_medium=email&utm_campaign=event-agent-sandboxes-live-demo&utm_content=launch-email-button )

Can't make it live? Register and we'll send you the recording.

The Pydantic team

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

You're receiving this because you opted in for Pydantic updates.

Unsubscribe ( https://track.customer.io/unsubscribe/dgSI0QoDAMmABMiABAGh0UiRGdgIGZoPLu3Mzgc= ) or
manage tracking preferences ( https://track.customer.io/tracking_consent/dgSI0QoDAMmABMiABAGh0UiRGdgIGZoPLu3Mzgc= )

GitHub ( https://github.com/pydantic/pydantic ) |
Website ( https://pydantic.dev?utm_source=newsletter&utm_medium=email&utm_campaign=launch-agent-sandboxes&utm_content=footer ) |
Community ( https://logfire.pydantic.dev/docs/help/ ) |
X ( https://x.com/pydantic ) |
LinkedIn ( https://www.linkedin.com/company/pydantic ) |
YouTube ( https://www.youtube.com/@Pydantic?utm_source=newsletter&utm_medium=email&utm_campaign=launch-agent-sandboxes&utm_content=footer )

Accession

nse/pydantic/2026-11-12/6828

From

Samuel Colvin · samuelcolvin@pydantic.dev

Sent

12 November 2026 · 19:00 UTC