<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Undreamt</title>
    <description>The latest articles on DEV Community by Undreamt (undreamt-oss).</description>
    <link>https://dev.to/undreamt-oss</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Forganization%2Fprofile_image%2F15770%2Fe8cdb45c-80fb-44c0-8ebd-1cf0106e45ff.png</url>
      <title>DEV Community: Undreamt</title>
      <link>https://dev.to/undreamt-oss</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/undreamt-oss"/>
    <language>en</language>
    <item>
      <title>Introducing Gabby — An Agent Runtime and Governance Framework</title>
      <dc:creator>Pranav Manglik</dc:creator>
      <pubDate>Tue, 17 Nov 2026 20:17:32 +0000</pubDate>
      <link>https://dev.to/undreamt-oss/introducing-gabby-an-agent-runtime-and-governance-framework-8p5</link>
      <guid>https://dev.to/undreamt-oss/introducing-gabby-an-agent-runtime-and-governance-framework-8p5</guid>
      <description>&lt;p&gt;Today, I'm introducing &lt;a href="https://github.com/undreamt-oss/gabby" rel="noopener noreferrer"&gt;Gabby&lt;/a&gt;, an open-source project I've been building to explore a different part of the AI agent ecosystem: what happens around an agent when it actually runs?&lt;/p&gt;

&lt;p&gt;There are plenty of frameworks focused on building agents, connecting them to tools, and orchestrating workflows. But I wanted to work on a different problem: &lt;strong&gt;how do we make agent execution more controlled, predictable, extensible, and auditable?&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;That's where Gabby comes in.&lt;/p&gt;

&lt;p&gt;Gabby is a &lt;strong&gt;stateless agent runtime and governance framework&lt;/strong&gt;. It provides the execution environment in which model-driven agents can operate while enforcing explicit boundaries around what they can do.&lt;/p&gt;

&lt;p&gt;Some of its core capabilities include:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Agent definitions:&lt;/strong&gt; Define agents through reusable, structured configurations.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Runtime governance:&lt;/strong&gt; Enforce tool permissions, policies, execution limits, and approval requirements.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Skills:&lt;/strong&gt; Create, compose, resolve, and reuse modular agent capabilities.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Provider flexibility:&lt;/strong&gt; Work with different model providers without tightly coupling the runtime to one ecosystem.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Knowledge and retrieval:&lt;/strong&gt; Integrate knowledge sources and retrieval mechanisms into agent execution.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Execution tracing:&lt;/strong&gt; Make agent activity more observable and auditable.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Multi-agent support:&lt;/strong&gt; Enable agent composition and host-orchestrated multi-agent workflows.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;One architectural decision is particularly important: &lt;strong&gt;Gabby is stateless by design.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;It does not attempt to own your application's conversations, persistent memory, or entire workflow. The host application remains responsible for durable state and orchestration. Gabby focuses on governing and executing agent tasks.&lt;/p&gt;

&lt;p&gt;It is also important to clarify what Gabby is not. It is not an LLM, does not train models, and is not intended to replace orchestration frameworks. It is designed to complement the existing ecosystem by providing a dedicated runtime and governance layer.&lt;/p&gt;

&lt;h3&gt;
  
  
  A realistic beginning
&lt;/h3&gt;

&lt;p&gt;Gabby is not perfect, and I don't want to present it as production-ready infrastructure without qualification.&lt;/p&gt;

&lt;p&gt;There are still limitations, architectural trade-offs, areas that need more testing, and decisions that will evolve as the project matures. The project is currently in beta, and its APIs and extension contracts may change.&lt;/p&gt;

&lt;p&gt;There is also a lot left to validate through real-world workloads, independent security reviews, performance testing, and feedback from developers who use it.&lt;/p&gt;

&lt;p&gt;I see this release as a starting point rather than a finish line.&lt;/p&gt;

&lt;h3&gt;
  
  
  Why open source?
&lt;/h3&gt;

&lt;p&gt;Because agent infrastructure should be something developers can inspect, understand, challenge, extend, and improve — not just consume as a black box.&lt;/p&gt;

&lt;p&gt;I'm releasing Gabby under the Apache-2.0 license and would genuinely appreciate feedback, criticism, contributions, and discussions around its architecture and direction.&lt;/p&gt;

&lt;p&gt;If you're interested in agent runtimes, execution governance, tool security, or multi-agent systems, I'd like you to take a look.&lt;/p&gt;

&lt;p&gt;Built in the open. Still evolving. Open to being challenged.&lt;/p&gt;

</description>
      <category>agents</category>
      <category>ai</category>
      <category>architecture</category>
      <category>opensource</category>
    </item>
  </channel>
</rss>
