<?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: David Cooper</title>
    <description>The latest articles on DEV Community by David Cooper (@dcooper).</description>
    <link>https://dev.to/dcooper</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%2Fuser%2Fprofile_image%2F4138210%2Fb44a56a3-c330-4d1b-b489-5d4788c1df20.jpg</url>
      <title>DEV Community: David Cooper</title>
      <link>https://dev.to/dcooper</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/dcooper"/>
    <language>en</language>
    <item>
      <title>I built BHive: a multi-agent framework for Gemini/OpenAI/Anthropic/Ollama. Tear it apart.</title>
      <dc:creator>David Cooper</dc:creator>
      <pubDate>Tue, 22 Sep 2026 19:33:28 +0000</pubDate>
      <link>https://dev.to/dcooper/i-built-bhive-a-multi-agent-framework-for-geminiopenaianthropicollama-tear-it-apart-2c8c</link>
      <guid>https://dev.to/dcooper/i-built-bhive-a-multi-agent-framework-for-geminiopenaianthropicollama-tear-it-apart-2c8c</guid>
      <description>&lt;p&gt;I've been building this for two weeks and want honest technical feedback before I invest more time.&lt;/p&gt;

&lt;h2&gt;
  
  
  What it does
&lt;/h2&gt;

&lt;p&gt;Bhive is a multi-agent framework in TypeScript. Agents ("Bees") can delegate tasks to each other, call tools, and it works with any LLM provider — Gemini, OpenAI, Anthropic, or local Ollama — through the same code.&lt;/p&gt;

&lt;h2&gt;
  
  
  Architecture decisions I'm unsure about
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Provider abstraction via an &lt;code&gt;LLMAdapter&lt;/code&gt; interface, with a &lt;code&gt;ToolCallingParser&lt;/code&gt; that normalizes tool-call formats across providers. Good pattern, or overengineered for what it does?&lt;/li&gt;
&lt;li&gt;Tool-call validation happens in the Bee's execution loop (not inside each adapter). Right layer for that?&lt;/li&gt;
&lt;li&gt;Timeout resolution has three levels (explicit arg → env var → default). Reasonable, or too much configuration surface?&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What I'd love feedback on
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;The security model (&lt;code&gt;ToolCallValidator&lt;/code&gt;: injection guards, allowlists, size limits) — anything missing?&lt;/li&gt;
&lt;li&gt;Whether the delegation pattern (Bee → Bee, via &lt;code&gt;createDelegationTool&lt;/code&gt;) makes sense as an abstraction&lt;/li&gt;
&lt;li&gt;Anything that smells wrong in the repo structure&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Repo: &lt;a href="https://github.com/dcooperdev/bhive-core" rel="noopener noreferrer"&gt;https://github.com/dcooperdev/bhive-core&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Tests: 299 passing, ~98% coverage. But test coverage isn't the same as "this is well-designed" — that's what I actually want checked.&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>opensource</category>
      <category>typescript</category>
      <category>llm</category>
    </item>
  </channel>
</rss>
