Wrap Claude Code, opencode, or any agent CLI of your choice. Run autonomous multi-agent workflows that gather data, analyze it, deliver it, and heal themselves when things break — on the LLM subscription you already pay for.
The Agent Office — agents working in real time
Chat interface — natural language task control
Type anything — agents understand intent, typos and all
Tasks get scheduled automatically when needed
Pipeline running — workers gather, analysts produce, validators check
Project complete — artifacts delivered
Project view — pipeline in progress
Project view — pipeline completed
Files & activity — every step audited
Full activity log — nothing gets lost
Scheduled jobs — agents discover their own cadence
Artifacts delivered — PDFs, structured data, API
Admin console — agent servers & API key management
Pipeline architecture
Component map
Phase types
OpenRed wraps the agentic coding CLI you already use — Claude Code, opencode, Gemini CLI, OpenAI Codex, Aider, Goose, or whatever drops next week. Around that, it adds the orchestration layer that none of those tools ship with.
OpenRed doesn't bill per token, take a markup on inference, or lock you into a particular provider. It drives the CLI of your choice using the subscription or API key you already own. Anthropic Max, DeepSeek API, local Ollama via opencode — pick what fits, swap when something better ships.
Currently running in production at Bimini Technologies, where it powers the Sonar Market Intelligence pipeline — fifty-plus daily cron jobs gathering, analyzing, and delivering financial data without human supervision.
OpenRed is an operations layer. AI agents that execute multi-step workflows, evaluate their own output, and recover from failures automatically.
Workers gather data. Analysts produce reports. Validators check quality. Each agent has a defined role, workspace, and skill set. Red orchestrates them all.
When a delivery fails, the system routes back to fix it. When the scheduler goes down, missed jobs are detected and queued on restart. No data loss, no silent failures.
Scale horizontally. Add remote agent servers, and the round-robin dispatcher distributes work across them. One goes down? Traffic flows to the rest.
"Get me the latest CPI data and schedule it based on their release calendar." That's it. Red figures out the rest — pipeline, scheduling, delivery.
Agents discover publication schedules and auto-create cron jobs. Post-completion checks verify your entire request was fulfilled before closing out.
Runs on your servers. Uses your LLM subscription or API key. No per-seat fees, no API markups, no data leaving your network.
Drive OpenRed from any application. Submit tasks, query project state, retrieve artifacts, manage agents — every action available in the UI is exposed as a clean REST endpoint.
OpenRed pushes events to your endpoints when jobs complete, validators flag issues, or artifacts are ready. Your downstream systems react in real time — no polling required.
You describe what you need. Red handles the rest.
Talk to Red through the chat interface. Describe what data you need, how you want it analyzed, and where it should go.
Red selects the right agents, assigns roles, and kicks off the pipeline. Workers fetch data. Analysts process it. Validators verify it.
Structured data, PDF reports, API deliveries. Red checks that your full request was satisfied — including scheduling if you asked for it.
OpenRed is currently a single-tenant production system. I'm evaluating whether to release it publicly.
The plumbing — the scheduler, the worker dispatch, the validator loop, the self-healing pipeline framework, the CLI-harness abstraction — is the kind of infrastructure plenty of teams have tried to build and most have given up on. I have it working, and it's saving real operational time at my first deployment. Releasing it could save a lot of duplicated effort across the agentic-tooling community.
What's still being weighed:
If you'd find OpenRed useful, want to contribute, or have thoughts on what the open source path should look like, I'd like to hear from you.
Drop me a note if you'd want to use it, contribute, or have feedback on the open source direction.
Get in Touch