The latest articles on DEV Community by member_0862620f (@member_0862620f). https://dev.to/member_0862620f 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%2F3228897%2Fcbe84779-4cb1-476e-bf48-dda496963c29.png https://dev.to/member_0862620f en member_0862620f Mon, 26 Oct 2026 06:01:41 +0000 https://dev.to/member_0862620f/i-built-a-local-first-context-fabric-for-ai-coding-agents-1pk https://dev.to/member_0862620f/i-built-a-local-first-context-fabric-for-ai-coding-agents-1pk <h1> I built a local-first context fabric for AI coding agents </h1> <p>If you use Claude Code, Codex CLI, or Kimi CLI across multiple projects, you already know the problem: every session starts from zero. Context is lost. It’s hard to see which repos are active, which are dead, and where an agent should actually work.</p> <p>I built <strong>Developer Control Plane</strong> to fix that.</p> <h2> The context-switch tax is real </h2> <p>I juggle 30+ projects. Some are in R&amp;D, some are half-built, some are production-ready. Switching between them means rebuilding mental context every time. Current AI coding agents make this worse: they’re amnesic between sessions. They forget architectural decisions, dependencies, and recent work.</p> <h2> Why current agents are amnesic </h2> <p>Claude Code, Codex CLI, and similar tools are stateless. Each run is a fresh conversation. They don’t remember:</p> <ul> <li>Which project you were working on yesterday</li> <li>What the last commit was</li> <li>Whether tests pass</li> <li>What stage the project is in</li> </ul> <p>You have to re-explain everything every time.</p> <h2> The three-layer architecture: signals → inferences → decisions </h2> <p>Developer Control Plane solves this with an event-sourced architecture built on three immutable layers:</p> <ol> <li> <strong>Raw Signals (Facts)</strong>: Deterministic data such as git commits, file modifications, test coverage presence, and open task counts.</li> <li> <strong>Inferences (Probabilities)</strong>: AI or heuristic-driven guesses about the project, such as assigning it a "Development" stage or calculating completion probability. Every inference includes a confidence score.</li> <li> <strong>Decisions (Actions)</strong>: System triggers, such as orchestrating an AI agent, suggesting refactors, or marking a project as dormant.</li> </ol> <p>This keeps the system honest. Instead of fake certainty, you get calibrated confidence that decays as a project sits idle.</p> <h2> How propose-only safety works </h2> <p>The biggest risk with AI coding agents is uncontrolled changes. Developer Control Plane uses a <strong>propose-only model</strong>:</p> <ol> <li>Agent runs are proposed, not auto-committed</li> <li>Changes stay uncommitted in your working tree</li> <li>You review the diff stat in the dashboard</li> <li>You <strong>approve</strong> to keep changes, or <strong>discard</strong> to revert</li> <li>Discard automatically reverts the exact files the run touched</li> </ol> <p>This means you never have to fear what an agent might break. Every change is reviewable and reversible.</p> <h2> Self-improving control plane </h2> <p>Here’s the part I’m most excited about: <strong>bottlenecks observed while agents work become tasks for the control plane itself</strong>.</p> <p>If an agent run fails, times out, or hits a guard skip, Developer Control Plane records it as a self-improvement task. A background loop executes at most one self-improvement task per day, and only when no other project’s work is pending.</p> <p>The control plane fixes its own bottlenecks.</p> <h2> Results after 2 weeks of use </h2> <ul> <li> <strong>Context recovery</strong>: I can resume any project in seconds, not minutes</li> <li> <strong>Visibility</strong>: The dashboard shows me which projects actually need work</li> <li> <strong>Safety</strong>: Zero “agent broke something and I didn’t notice” incidents</li> <li> <strong>Time savings</strong>: Probably 5+ hours per week on context-switching</li> </ul> <h2> How to try it </h2> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code>git clone https://github.com/Ishuin/Developer-Plane.git <span class="nb">cd </span>Developer-Plane python <span class="nt">-m</span> venv venv <span class="nb">source </span>venv/bin/activate <span class="c"># Windows: venv\Scripts\activate</span> pip <span class="nb">install</span> <span class="nt">-r</span> requirements.txt python src/main.py <span class="nt">--scan</span> D:<span class="se">\P</span>rojects python src/main.py <span class="nt">--host</span> 127.0.0.1 <span class="nt">--port</span> 8787 <span class="c"># Open http://127.0.0.1:8787</span> </code></pre> </div> <p>Or with Docker:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code>docker build <span class="nt">-t</span> developer-plane <span class="nb">.</span> docker run <span class="nt">-p</span> 8787:8787 <span class="nt">-v</span> %cd%:/app developer-plane </code></pre> </div> <h2> What’s next </h2> <ul> <li>Team sync and multi-user support</li> <li>Cloud backup for context fabric</li> <li>Plugin system for custom agents</li> <li>Advanced analytics and reporting</li> </ul> <p>If you’re juggling multiple AI coding agents, this might save you hours of context-switching per week.</p> <p>GitHub: <a href="https://github.com/Ishuin/Developer-Plane" rel="noopener noreferrer">https://github.com/Ishuin/Developer-Plane</a><br><br> License: AGPL-3.0</p> agents ai coding llm