agent-apprenticeship

本页保存的是公开项目资料快照，阅读过程不需要连接 GitHub。

Agent Apprenticeship

图片：npm version 图片：HF Downloads

Real-world agent work experience, looped into collective learning.

The living ecosystem where AI agents complete tasks through workflow loops, improve through iterative execution, are evaluated by mentor agents or humans in the loop, and turn completed work into reusable work experience and data to improve future agents.

npx agent-apprenticeship init

图片：Agent Apprenticeship

As agents move into long-horizon, economically valuable work, Agent Apprenticeship creates the open infrastructure where real-world tasks generate reusable learning signals and complex workflows advance through agent loops that turn execution into shared improvement.

Agent Apprenticeship is designed for a compounding exchange of agent work experience: economically valuable task execution generates training signals, those signals improve future work, and future work creates new reusable experience for the ecosystem.

Agent Apprenticeship is built for iterative workflow loops across domains, from simple tasks to complex specialized work. Apprentice agents work with mentor agents, users, or human experts to complete real-world tasks, while each workflow generates reusable learning signals for the ecosystem.

The latest seed dataset includes:

• 500+ curated seed tasks sourced and grounded from the real world
• 495 reusable agent lessons
• 1000+ full agent execution traces
• 1000+ agent work episodes / task rollouts
• 505 full agent work experience compilations
• 39k+ structured experience compilation records

The seed dataset spans specialized, economically valuable tasks across domains and forms the first layer of the Agent Apprenticeship ecosystem.

Agent Apprenticeship is now available for anyone to start using with local agents, including Codex, Cursor, Claude Code, OpenClaw, OpenCode, Hermes Agent, and custom agents, alongside different model providers. Users can run automated agent workflow loops locally, contribute agent learning signals back to the ecosystem, and use shared ecosystem signals to improve their own agents.

Agent Apprenticeship is about the future of work and the economic value of agents. For every task executed through Agent Apprenticeship, the system can estimate task-level economic value, especially across specialized domains. It is built for everyday use to improve agent performance and outcome quality, while enabling users to exchange agent work experience with each other and with domain-expert-led agents in one living ecosystem.

Install

npx agent-apprenticeship init

Or install globally:

npm install -g agent-apprenticeship
apprentice init

The installed command is:

apprentice

The long-form command also remains available:

agent-apprenticeship

Quickstart

Start Agent Apprenticeship:

npx agent-apprenticeship init

Use defaults:

npx agent-apprenticeship init --defaults

Check your setup:

apprentice settings
apprentice doctor

Configure your Apprentice Agent, Mentor Model Provider, and Apprenticeship Mode:

apprentice configure
apprentice configure model
apprentice settings

Apprenticeship Modes:

Autonomous
Expert-Led
Organization Custom

Store Mentor Model Provider keys in:

~/.agent-apprenticeship/.env.local

Example:

OPENAI_API_KEY=""
ANTHROPIC_API_KEY=""
GEMINI_API_KEY=""
OPENROUTER_API_KEY=""

You can also use shell environment variables for the current terminal session:

export OPENAI_API_KEY="..."
apprentice doctor

Run your first task:

apprentice run "Create a short market map for AI procurement tools."

Watch progress:

apprentice watch 

When the run completes, Agent Apprenticeship prints the local run folder and Experience Compilation path.

Inspect the generated Experience Compilation:

apprentice ecosystem inspect 
apprentice bundle check 

Export the Full Experience Compilation:

apprentice ecosystem export  --full

Install Runtime Training from prior experience:

apprentice learn install 

Use installed Runtime Training in a future run:

apprentice run "Create a release checklist for an AI agent project."

Configure maximum loop depth:

apprentice settings

For a one-off terminal session:

export AA_MAX_ITERATIONS=3

Apprentice Agents

Available Apprentice Agents:

Codex
Cursor
Claude Code
OpenClaw
OpenCode
Hermes Agent
Custom

Agent Apprenticeship auto-detects installed CLIs. If multiple are detected, choose one during setup.

Custom lets you provide a command template:

apprentice configure agent custom --command-template "my-agent run --workspace {workspace} --prompt-file {prompt_file}"

Mentor Model Providers

Store local keys in:

~/.agent-apprenticeship/.env.local

Example:

OPENAI_API_KEY=""
ANTHROPIC_API_KEY=""
GEMINI_API_KEY=""
OPENROUTER_API_KEY=""

Configure:

apprentice configure model
apprentice doctor

Public Ecosystem and Private Internal Only

Configure contribution mode:

apprentice settings

Modes:

Public Ecosystem
Private Internal Only

Explore ecosystem experience:

apprentice ecosystem search 
apprentice ecosystem inspect 
apprentice ecosystem pull 

Export Full Experience Compilation:

apprentice ecosystem export  --full

Install Runtime Training:

apprentice learn install 

Seed Dataset v0.2

Seed Dataset v0.2 is available on Hugging Face:

https://huggingface.co/datasets/Forsy-AI/agent-apprenticeship-seed-dataset_v0.2

It includes 505 full agent work experience compilations and 39k+ structured experience compilation records.

Selected ALE Tasks Demo

Selected ALE workflow-family mirrored transfer tasks are included under:

selected_ale_tasks_demo/

Public Repo Structure

bin/
src/
schemas/
examples/
selected_ale_tasks_demo/