The latest articles on DEV Community by umut erturk (@umuterturk). https://dev.to/umuterturk 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%2F4224400%2F77b5e950-0ed5-4b64-9935-2bfb3333ce20.jpg https://dev.to/umuterturk en umut erturk Tue, 03 Nov 2026 14:14:58 +0000 https://dev.to/umuterturk/your-agent-doesnt-need-a-codebase-memory-1dni https://dev.to/umuterturk/your-agent-doesnt-need-a-codebase-memory-1dni <p>Every coding agent needs to understand your codebase. But does it really need to <em>remember</em> it?</p> <p>Most code intelligence tools maintain indexes, embeddings, or knowledge graphs, depends on other stuff like sqlite. These need to stay synchronized with your code, or risk giving agents outdated answers.</p> <p>I built <a href="https://github.com/umuterturk/cxj" rel="noopener noreferrer">cxj</a> to try a simpler approach: <strong>read the current source, answer the question, and forget everything.</strong></p> <h2> What is cxj? </h2> <p>An open-source, stateless Java/Spring code intelligence CLI for coding agents.</p> <p>Instead of spending tokens exploring dozens of files, your agent gets structured answers in milliseconds.<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code>cxj brief <span class="c"># Understand the entire service</span> cxj endpoints <span class="c"># List HTTP routes</span> cxj trail PaymentService.pay <span class="c"># Follow calls and find tests</span> cxj callers PaymentService.pay <span class="c"># Find callers</span> cxj deps <span class="c"># Get resolved dependency versions</span> </code></pre> </div> <p>It understands Spring-specific concepts, resolves method calls using declared types, and explicitly reports uncertainty instead of silently guessing.</p> <h2> Does it actually help? </h2> <p>I compared cxj against traditional grep + file exploration using the same AI model on a production Spring service.</p> <p>For six structural questions, both approaches answered correctly:</p> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th></th> <th>Improvement with cxj</th> </tr> </thead> <tbody> <tr> <td>Input tokens</td> <td><strong>6.4× fewer</strong></td> </tr> <tr> <td>Agent cost</td> <td><strong>5.6× cheaper</strong></td> </tr> <tr> <td>Completion time</td> <td><strong>2.2× faster</strong></td> </tr> </tbody> </table></div> <p>For simple one-line lookups, <strong>grep actually won</strong>. cxj is most useful when answers span multiple files.</p> <p><a href="https://github.com/umuterturk/cxj/blob/main/docs/MEASURING.md" rel="noopener noreferrer">Full benchmark and methodology</a>.</p> <h2> No index. No daemon. No stale cache. </h2> <p>cxj uses tree-sitter to parse the current code on every invocation, builds its knowledge in memory, and exits.</p> <ul> <li>561-file production Spring service: <strong>~170 ms</strong> </li> <li>Spring Framework, 1.5 million lines: <strong>2.4 seconds</strong> </li> </ul> <p>No persistent state, no background process, and no network required for queries.</p> <h2> Try it </h2> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code>brew <span class="nb">install </span>umuterturk/tap/cxj npx skills add umuterturk/cxj <span class="o">[</span>download] <span class="o">(</span>https://github.com/umuterturk/cxj/releases/latest<span class="o">)</span> </code></pre> </div> <p>Works with Claude Code, Codex, OpenCode, and other skill-compatible agents. macOS and Linux. MIT licensed.</p> <p><strong>GitHub:</strong> <a href="https://github.com/umuterturk/cxj" rel="noopener noreferrer">https://github.com/umuterturk/cxj</a><br> Come contribute!</p> <p>I'm curious whether other developers find persistent codebase indexes necessary, or whether parsing on demand is already fast enough for most projects.</p> ai showdev java springboot