SeedX

What SeedX does

SeedX is a multi-agent system that turns a work or learning question into a learning package that can be executed, evaluated and transferred. A user sends +ask from Claude Code, Hermes, OpenClaw or another connected entry point. A hook saves the question under input/questions/, starts the orchestrator, opens a visualizer and writes the results to an output folder named after the topic and a timestamp.

A question-planner creates a learning plan, contract and design guide, then three fixed tasks run in order: frame the question, build the mastery path with exercises and checkpoints, and apply and transfer the result. After each task a learning-evaluator passes or fails the work, and a failure resumes the same builder. Each run leaves deliverables, agent notes and a run log with events and state. The project also serves as a testbed for long, unattended agent runs, including experiments with Chinese-model agents.

Key features

• +ask and +start triggers from Claude Code, Hermes, OpenClaw and other connected agents
• Question-planner, mastery-builder and learning-evaluator roles
• Three fixed tasks: frame the question, build the mastery path, apply and transfer
• Evaluator pass or fail with repair of weak parts
• Visualizer showing task progress, agent handoffs and the event timeline
• Output folder with deliverables, agent notes, events.jsonl and state.json

When to use it

• Turning a messy work question into a step-by-step learning path with exercises
• Testing whether model agents can finish long multi-step work with no human in the loop
• Sending a question from a phone-based agent workflow and receiving a full package

Who it is for: Learners who want a structured study plan from a question, and agent builders testing long unattended runs.

How it fits with Hermes Agent

It treats Hermes as one of several connected agent entry points that can trigger a run; it is not built only for Hermes Agent.

FAQ