Ampere Agent
Ampere Agent is a main-agent orchestration system that distributes and schedules tasks for execution by external agents such as Codex and Claude Code. It dispatches work to these subordinate agents, coordinates their execution, collects the results in a unified manner, and delivers the consolidated output.
Timeline (last 90 days)
In the eleventh episode of the "Build Your Own Agent from Scratch" series, Bilibili UP host Xiaoshan demonstrated a complete solution in which the main Agent (Ampere Agent) schedules Codex and Claude Code to complete complex tasks.
The solution designs two return paths: process progress (public text, tool status, etc. organized by the adapter into unified events) and completion notification (results are collected and a notification is sent when the task ends, and the main process decides whether to resume or deliver).
The complete scheduling flow includes: the user dispatches a task, the main process breaks it down into work packages, selects an executor based on capability and dispatches, the executor returns results, the main process checks whether the standards are met (if not, it adds requirements and resumes the original task), and delivers the results.
The UP host actually demonstrated a task: Ampere Agent first dispatched Codex to research the Tesla ramp car jump video and extract key process images, then dispatched Claude Code to build a 3D Tesla car jump demonstration video reproducing it, and the task lasted several hours.
The project's architecture (adapter layer, delegation service, dual-channel return, and the check-resend-resume closed loop) does not depend on any particular product and can be migrated to other combinations of Agents.
All Facts (5)
In the eleventh episode of the "Build Your Own Agent from Scratch" series, Bilibili UP host Xiaoshan demonstrated a complete solution in which the main Agent (Ampere Agent) schedules Codex and Claude Code to complete complex tasks.
50%UnverifiedThe solution designs two return paths: process progress (public text, tool status, etc. organized by the adapter into unified events) and completion notification (results are collected and a notification is sent when the task ends, and the main process decides whether to resume or deliver).
50%UnverifiedThe complete scheduling flow includes: the user dispatches a task, the main process breaks it down into work packages, selects an executor based on capability and dispatches, the executor returns results, the main process checks whether the standards are met (if not, it adds requirements and resumes the original task), and delivers the results.
50%UnverifiedThe project's architecture (adapter layer, delegation service, dual-channel return, and the check-resend-resume closed loop) does not depend on any particular product and can be migrated to other combinations of Agents.
50%UnverifiedThe UP host actually demonstrated a task: Ampere Agent first dispatched Codex to research the Tesla ramp car jump video and extract key process images, then dispatched Claude Code to build a 3D Tesla car jump demonstration video reproducing it, and the task lasted several hours.
50%