Developer builds self-hosted AI agent runtime to keep execution off third-party clouds

·1 views

A developer has launched VaultRun, an open-source, self-hosted runtime designed to give engineers direct control over where AI agents execute code. The project was motivated by concerns that hosted sandboxes are unsuitable when agents handle credentials, customer data, or production infrastructure. VaultRun runs each agent session in an isolated Docker container with network access disabled by default, requiring explicit permission grants for any connectivity. It also logs agent actions in a signed audit trail, allowing teams to reconstruct exactly what commands were run, when, and with what results. Released under the Apache 2.0 license, the project is in early development and available on GitHub, with the creator inviting feedback from developers navigating similar security tradeoffs.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

New AWS S3 Annotations Feature Has a Security Gap That Bypasses Integrity Checks

A security researcher demonstrated that AWS S3's newly launched annotations feature, released in June 2026, can be exploited to alter a file's metadata context without modifying the file itself. In a proof-of-concept, a HIPAA-regulated health record stored in S3 had its security label changed to 'public' purely by editing its annotation, leaving the file's checksum completely unchanged. Standard integrity monitors scanning the storage bucket detected no tampering, since the underlying object was never touched. The vulnerability stems from annotations being independently writable and mutable, meaning permission to edit context metadata is a separate, potentially overlooked access control. The researcher argues this creates a significant blind spot in data classification pipelines built on top of the new feature, which supports up to 1 GB of structured metadata per object and is queryable via Amazon Athena.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Java 21 Virtual Threads Boost Concurrency but Can Overwhelm Database Connection Pools

Java 21 introduced virtual threads via Project Loom, allowing applications to handle tens of thousands of simultaneous requests without manual thread-pool tuning. However, this surge in concurrency can expose a hidden bottleneck: fixed-size database connection pools such as HikariCP, which were originally sized for lower concurrency levels. When thousands of virtual threads compete for a pool of only a few dozen connections, requests queue up and time out, shifting the crash from the compute layer to the data layer. Previously, platform threads inadvertently capped concurrency and acted as an unintentional safeguard for downstream resources. Developers adopting virtual threads are advised to reassess and right-size their connection pools based on what the database can actually sustain, not merely what the application demands.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Stripe Acquires AI Model Router OpenRouter for Over $7 Billion

Payments giant Stripe has finalized a deal to acquire OpenRouter, an AI model routing platform, for more than $7 billion, Bloomberg reported on August 16. Founded in 2023 by Alex Atallah, OpenRouter provides a single API giving developers access to over 400 AI models from providers including OpenAI, Anthropic, and Google, charging a 5% fee on inference spend. The acquisition price represents more than five times OpenRouter's $1.3 billion Series B valuation, which was set just 82 days earlier in late May 2026. Stripe and OpenRouter had already been working closely together since January 2026, when OpenRouter adopted Stripe's invoicing, tax, and fraud tools to manage its global billing. Analysts see the deal as a logical extension of Stripe's core expertise in routing high-volume, latency-sensitive transactions across multiple providers — a structure closely mirrored by AI inference routing.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Python LEGB Scope, First-Class Functions, and Common Pitfalls Explained

A new developer guide on DEV Community breaks down several intermediate Python concepts that often confuse beginners. The article explains Python's LEGB scoping rule, which defines the order — Local, Enclosing, Global, and Built-in — in which Python resolves variable names. It also covers the use of the nonlocal keyword, which allows inner functions to modify variables belonging to an enclosing function's scope. Additional topics include first-class functions, flexible argument passing with *args and **kwargs, and the risks of using mutable objects as default function arguments. The guide aims to make these concepts approachable through simple definitions and step-by-step code examples.

0 comments Read more at DEV Community