Launching today
BDFL - Benevolent Delegator for LLMs
Plan deliberately. Build in parallel. Stay in control.
2 followers
Plan deliberately. Build in parallel. Stay in control.
2 followers
BDFL is an open-source terminal supervisor for Codex, Claude Code, and Ollama sessions. Work with a planning agent, compare and approve versioned plans or individual sections of it, then let isolated worker agents implement the approved work while BDFL handles scheduling, checks, review, verification, integration, and recovery.
Hi everyone👋 I kept running into this problem, where I would ask an agent for a small change in plan, get a completely rewritten plan back, and then be too tired to read the whole thing again.
I wanted a way to see exactly what changed in the plan, review smaller sections, lock sessions so they don't change upon revision and go back and execute an older version when my latest instructions somehow made everything worse.
I wanted to execute plans in smaller chunks without filling up my main terminal window, make chunks run through subagents that I can still interact with and be able to change direction for particular subagent without restarting the whole workflow
That became BDFL: the open-source delegator, and not the dictator.
You can do everything I mentioned above. It works with @OpenAI Codex CLI, @Claude Code, and also supports @Ollama models. You can plan with one model, execute with another, mix open and closed models, or arrange the workflow however you prefer.
PS: It's OpenAI day, so wanted to launch it today. The workflow is still fresh. Feedback, bug reports, ideas, stars, and contributions would mean a lot.