In early 2026, I set out to automate market intelligence workflows for Bimini Technologies. The task was straightforward: gather financial data from widely disparate sources, analyze it, deliver it, and do it on a schedule that fluctuates per source - making it incredibly difficult to keep up with. Every week. Every month. Without someone babysitting it.

I tried OpenClaw. Huge letdown. The promise of an open-source agentic framework sounded perfect on paper. In practice? Brittle pipelines, no orchestration layer, and the moment Anthropic released their own managed agent service, the project was effectively dead. Until OpenAI acqui-hired the creator of OpenClaw and resurrected the project - yet still falls short of being capable of extreme automation.

I tried Hermes Agent. The "self-learning" pitch was compelling. In reality? The self-learning bit is a nothing burger. It's a wrapper around prompt templates with a feedback loop that doesn't meaningfully improve over time. And it was nowhere near as automated as it should be - you still had to hand-hold every step.

I looked at Claude Managed Agents from Anthropic. Real agents, real tool use, real capability. But no pre-built orchestration. No evaluation loop. No scheduling. No self-healing. You'd still need self-healing, extreme automation potential to build what I needed, and the per-run costs at scale ran far above what a flat Claude subscription would have charged for the same work.

So I built OpenRed. From scratch. A harness that wraps the agentic CLI of your choice - Claude Code, opencode, Gemini CLI, OpenAI Codex, Aider, Goose, or whatever drops next week - and rides on the subscription or API key you already pay for. On top of that I add everything the market is missing: an orchestration layer (Red), specialized agent roles, pipeline-based workflows with evaluation loops, distributed worker servers, self-healing scheduling, and a clean interface to control it all. Plug in the model you trust; swap it tomorrow if a better one ships.

OpenRed isn't a SaaS dashboard. It's the infrastructure underneath one. Custom workflows, custom agent configurations, deployed on your servers. You own everything. No lock-in. No platform fees. No data leaving your network.

My first production deployment is Bimini.one, where OpenRed powers the Sonar Market Intelligence platform - gathering economic data, analyzing it through custom-trained AI models, correlating it against proprietary data through automated workflows, and delivering it on schedule to the Sonar interface API. 102+ daily cron-fired jobs, multi-agent pipelines, self-healing on failure. It's been running quietly in production since spring 2026.

Open source consideration

Now that OpenRed has earned its production stripes, I'm weighing whether to release it publicly. The plumbing is rare enough that releasing it could save a lot of duplicated effort across teams trying to build agentic infrastructure on top of CLIs like Claude Code and opencode. I'm working through licensing, maintenance commitment, and whether a small managed-hosting offer would make sense alongside the open source release for people who want the orchestration layer without operating it themselves.

If you'd find OpenRed useful, want to contribute, or have an opinion on what shape the open source release should take - I'd like to hear from you.