Elephant Agent Personal AI Skill

Elephant Agent 个人AI技能

Skill by ara.so — AI Agent Skills collection.

Elephant Agent is a Personal-Model First Self Evolving AI Agent that starts from understanding the person, not the task. It builds a correctable Personal Model across four lenses (Identity, World, Pulse, Journey) and helps design living Paths for work, health, habits, learning, and other long-running directions.

该技能来自ara.so —— AI Agent技能合集。

Elephant Agent是一款以个人模型为核心的自主进化AI Agent，它从了解用户本身而非任务开始。它通过四个维度（Identity、World、Pulse、Journey）构建可修正的Personal Model，并帮助为工作、健康、习惯、学习及其他长期方向设计动态的Paths。

What Elephant Agent Does

Elephant Agent 功能介绍

Personal Model: Builds understanding of who you are (Identity), what surrounds you (World), what's alive right now (Pulse), and what your path has taught (Journey)
Paths: Long-running arcs that break down into Steps with Checkpoints for human judgment
Herd Management: Coordinates Mother elephant and baby elephants under one understanding
Skills & Tools: Extensible system for browser, filesystem, MCP, and operator actions
Multi-Surface: Native macOS app + CLI + Dashboard for different workflows

Personal Model：构建对你是谁（Identity）、所处环境（World）、当前状态（Pulse）以及成长历程（Journey）的认知
Paths：长期任务轨迹，可拆解为带人工审核检查点的步骤（Steps）
Herd Management：在统一认知下协调母象和小象Agent
Skills & Tools：可扩展系统，支持浏览器、文件系统、MCP及操作员操作
Multi-Surface：原生macOS应用 + CLI + 控制面板，适配不同工作流

Installation

安装步骤

macOS Desktop App (Recommended)

macOS桌面应用（推荐）

bash

undefined

bash

undefined

Download from GitHub releases

curl -L -o elephant-agent.dmg https://github.com/agentic-in/elephant-agent/releases/latest/download/Elephant-Agent.dmg

Or visit releases page

open https://github.com/agentic-in/elephant-agent/releases/latest

undefined

open https://github.com/agentic-in/elephant-agent/releases/latest

undefined

CLI + Dashboard (Linux/Cloud/SSH)

CLI + 控制面板（Linux/云服务器/SSH）

bash

undefined

bash

undefined

Install via install script

curl -fsSL https://elephant.agentic-in.ai/install.sh | bash

Or manual installation

git clone https://github.com/agentic-in/elephant-agent.git cd elephant-agent pip install -e .

undefined

git clone https://github.com/agentic-in/elephant-agent.git cd elephant-agent pip install -e .

undefined

Key CLI Commands

核心CLI命令

Initialization and Setup

初始化与配置

bash

undefined

bash

undefined

Initialize Elephant Agent (first run)

elephant init

Check system readiness

elephant status

Configure provider settings

elephant config set provider openai elephant config set model gpt-4 elephant config set curiosity_effort medium

List all configuration

elephant config list

undefined

elephant config list

undefined

Daily Interaction

日常交互

bash

undefined

bash

undefined

Enter the chat TUI (main interaction mode)

elephant wake

Quick query without entering TUI

elephant ask "What should I focus on today?"

View Personal Model

elephant model show

View specific model lens

elephant model show --lens identity elephant model show --lens world elephant model show --lens pulse elephant model show --lens journey

undefined

elephant model show --lens identity elephant model show --lens world elephant model show --lens pulse elephant model show --lens journey

undefined

Paths Management

Paths管理

bash

undefined

bash

undefined

List all paths

elephant paths list

Create a new path

elephant paths create "Launch new product feature"

View path details

elephant paths show <path-id>

Update path status

elephant paths update <path-id> --status active

Archive completed path

elephant paths archive <path-id>

undefined

elephant paths archive <path-id>

undefined

Herd Management

Herd管理

bash

undefined

bash

undefined

List all agents in herd

elephant herd list

Create a baby elephant for specific task

elephant herd spawn --role researcher --context "market analysis"

View agent details

elephant herd show <agent-id>

Remove agent from herd

elephant herd remove <agent-id>

undefined

elephant herd remove <agent-id>

undefined

Skills and Tools

技能与工具

bash

undefined

bash

undefined

List available skills

elephant skills list

Enable a skill

elephant skills enable filesystem

Disable a skill

elephant skills disable browser

View skill documentation

elephant skills info filesystem

undefined

elephant skills info filesystem

undefined

Dashboard

控制面板

bash

undefined

bash

undefined

Open dashboard in browser

elephant dashboard

Run dashboard without opening browser (for remote/SSH)

elephant dashboard --no-open

Dashboard on custom port

elephant dashboard --port 8080

undefined

elephant dashboard --port 8080

undefined

Configuration

配置说明

Provider Configuration

服务商配置

bash

undefined

bash

undefined

OpenAI

elephant config set provider openai export OPENAI_API_KEY=your-key-here

Anthropic

elephant config set provider anthropic export ANTHROPIC_API_KEY=your-key-here

Local models (Ollama)

elephant config set provider ollama elephant config set model llama2

undefined

elephant config set provider ollama elephant config set model llama2

undefined

Configuration File

配置文件

Elephant Agent stores configuration in

~/.elephant/config.yaml

:

yaml

undefined

Elephant Agent将配置存储在

~/.elephant/config.yaml

：

yaml

undefined

~/.elephant/config.yaml

provider: openai model: gpt-4 curiosity_effort: medium # low, medium, high language: en boundaries:

respect_privacy
ask_before_destructive
explain_reasoning posture: collaborative # directive, collaborative, suggestive

undefined

provider: openai model: gpt-4 curiosity_effort: medium # low, medium, high language: en boundaries:

respect_privacy
ask_before_destructive
explain_reasoning posture: collaborative # directive, collaborative, suggestive

undefined

Environment Variables

环境变量

bash

undefined

bash

undefined

Provider API keys

export OPENAI_API_KEY=sk-... export ANTHROPIC_API_KEY=sk-ant-... export DEEPSEEK_API_KEY=...

Elephant-specific settings

export ELEPHANT_HOME=~/.elephant export ELEPHANT_LOG_LEVEL=info export ELEPHANT_DASHBOARD_PORT=3000

undefined

export ELEPHANT_HOME=~/.elephant export ELEPHANT_LOG_LEVEL=info export ELEPHANT_DASHBOARD_PORT=3000

undefined

Python API Usage

Python API使用

Basic Interaction

基础交互

python

from elephant_agent import ElephantAgent, PersonalModel

python

from elephant_agent import ElephantAgent, PersonalModel

Initialize agent

agent = ElephantAgent( provider="openai", model="gpt-4", curiosity_effort="medium" )

Load or create Personal Model

model = agent.get_personal_model()

Ask a question

response = agent.ask("What should I prioritize today?") print(response.message) print(f"Confidence: {response.confidence}")

undefined

response = agent.ask("What should I prioritize today?") print(response.message) print(f"Confidence: {response.confidence}")

undefined

Working with Personal Model

操作Personal Model

python

from elephant_agent import PersonalModel

python

from elephant_agent import PersonalModel

Access model lenses

model = PersonalModel.load()

View identity

identity = model.get_lens("identity") print(f"Values: {identity.values}") print(f"Decision style: {identity.decision_style}")

Update world lens

model.update_lens("world", { "people": ["Alice (mentor)", "Bob (colleague)"], "projects": ["Product launch", "Team onboarding"], "tools": ["VS Code", "Linear", "Slack"] })

Check pulse

pulse = model.get_lens("pulse") print(f"Current focus: {pulse.focus}") print(f"Energy level: {pulse.energy}") print(f"Constraints: {pulse.constraints}")

Save changes

model.save()

undefined

model.save()

undefined

Creating and Managing Paths

创建与管理Paths

python

from elephant_agent import Path, Step

python

from elephant_agent import Path, Step

Create a new path

path = Path.create( title="Launch API v2", description="Ship new API with auth and webhooks", context=model )

Add steps

path.add_step(Step( title="Design API schema", description="Define endpoints, auth flow, webhook events", checkpoint=True # Requires human review ))

path.add_step(Step( title="Implement core endpoints", description="Build CRUD operations with auth", checkpoint=False ))

path.add_step(Step( title="Design API schema", description="Define endpoints, auth flow, webhook events", checkpoint=True # Requires human review ))

path.add_step(Step( title="Implement core endpoints", description="Build CRUD operations with auth", checkpoint=False ))

Start the path

path.start()

Get next step

next_step = path.get_next_step() print(f"Next: {next_step.title}")

Mark step complete

path.complete_step(next_step.id)

Save path

path.save()

undefined

path.save()

undefined

Working with the Herd

操作Herd

python

from elephant_agent import Herd, BabyElephant

python

from elephant_agent import Herd, BabyElephant

Get the herd

herd = Herd.load()

Spawn a baby elephant for a specific task

researcher = BabyElephant( role="researcher", context={ "focus": "competitor analysis", "constraints": ["public data only"], "reporting_to": "mother" } )

herd.add(researcher)

researcher = BabyElephant( role="researcher", context={ "focus": "competitor analysis", "constraints": ["public data only"], "reporting_to": "mother" } )

herd.add(researcher)

Assign task to baby

task = researcher.assign_task( "Research top 3 competitors' API offerings" )

Check status

status = researcher.get_status() print(f"Progress: {status.progress}%")

Get results

if task.is_complete(): results = task.get_results() print(results.summary)

undefined

if task.is_complete(): results = task.get_results() print(results.summary)

undefined

Skills Integration

技能集成

python

from elephant_agent.skills import FileSystemSkill, BrowserSkill

python

from elephant_agent.skills import FileSystemSkill, BrowserSkill

Initialize skills

fs_skill = FileSystemSkill( allowed_paths=["/home/user/projects"], readonly=False )

browser_skill = BrowserSkill( headless=True, timeout=30 )

fs_skill = FileSystemSkill( allowed_paths=["/home/user/projects"], readonly=False )

browser_skill = BrowserSkill( headless=True, timeout=30 )

Register skills with agent

agent.register_skill(fs_skill) agent.register_skill(browser_skill)

Use skills in context

response = agent.ask( "Read the README.md and create a summary document", skills=["filesystem"] )

undefined

response = agent.ask( "Read the README.md and create a summary document", skills=["filesystem"] )

undefined

Advanced: Custom Skills

进阶：自定义技能

python

from elephant_agent.skills import Skill, SkillParameter

class GitSkill(Skill):
    name = "git"
    description = "Execute git commands safely"
    
    parameters = [
        SkillParameter("command", str, "Git command to run"),
        SkillParameter("args", list, "Command arguments", required=False)
    ]
    
    def execute(self, command: str, args: list = None):
        """Execute git command with safety checks."""
        import subprocess
        
        # Safety: only allow read operations
        safe_commands = ["status", "log", "diff", "branch"]
        if command not in safe_commands:
            return {
                "success": False,
                "error": f"Command '{command}' not allowed"
            }
        
        cmd = ["git", command] + (args or [])
        result = subprocess.run(
            cmd,
            capture_output=True,
            text=True
        )
        
        return {
            "success": result.returncode == 0,
            "output": result.stdout,
            "error": result.stderr
        }

python

from elephant_agent.skills import Skill, SkillParameter

class GitSkill(Skill):
    name = "git"
    description = "Execute git commands safely"
    
    parameters = [
        SkillParameter("command", str, "Git command to run"),
        SkillParameter("args", list, "Command arguments", required=False)
    ]
    
    def execute(self, command: str, args: list = None):
        """Execute git command with safety checks."""
        import subprocess
        
        # Safety: only allow read operations
        safe_commands = ["status", "log", "diff", "branch"]
        if command not in safe_commands:
            return {
                "success": False,
                "error": f"Command '{command}' not allowed"
            }
        
        cmd = ["git", command] + (args or [])
        result = subprocess.run(
            cmd,
            capture_output=True,
            text=True
        )
        
        return {
            "success": result.returncode == 0,
            "output": result.stdout,
            "error": result.stderr
        }

Register custom skill

git_skill = GitSkill() agent.register_skill(git_skill)

undefined

git_skill = GitSkill() agent.register_skill(git_skill)

undefined

Common Patterns

常见使用模式

Daily Check-in Pattern

每日签到模式

python

from elephant_agent import ElephantAgent
from datetime import datetime

agent = ElephantAgent.load()

python

from elephant_agent import ElephantAgent
from datetime import datetime

agent = ElephantAgent.load()

Morning routine

def morning_checkin(): model = agent.get_personal_model()

# Update pulse
model.update_pulse({
    "timestamp": datetime.now(),
    "energy": "high",
    "focus": ["API launch", "team sync"],
    "constraints": ["3hr meeting block 2-5pm"]
})

# Get prioritized guidance
guidance = agent.ask(
    "Given my current pulse and active paths, what should I prioritize today?"
)

return guidance

checkin = morning_checkin() print(checkin.message)

undefined

def morning_checkin(): model = agent.get_personal_model()

# Update pulse
model.update_pulse({
    "timestamp": datetime.now(),
    "energy": "high",
    "focus": ["API launch", "team sync"],
    "constraints": ["3hr meeting block 2-5pm"]
})

# Get prioritized guidance
guidance = agent.ask(
    "Given my current pulse and active paths, what should I prioritize today?"
)

return guidance

checkin = morning_checkin() print(checkin.message)

undefined

Path Progress Review

Path进度回顾

python

from elephant_agent import Path

def review_active_paths():
    paths = Path.list(status="active")
    
    for path in paths:
        print(f"\n=== {path.title} ===")
        print(f"Progress: {path.progress}%")
        
        next_step = path.get_next_step()
        if next_step:
            print(f"Next: {next_step.title}")
            
            if next_step.checkpoint:
                print("⚠️  Requires your review")
                
        blockers = path.get_blockers()
        if blockers:
            print(f"Blockers: {', '.join(blockers)}")

python

from elephant_agent import Path

def review_active_paths():
    paths = Path.list(status="active")
    
    for path in paths:
        print(f"\n=== {path.title} ===")
        print(f"Progress: {path.progress}%")
        
        next_step = path.get_next_step()
        if next_step:
            print(f"Next: {next_step.title}")
            
            if next_step.checkpoint:
                print("⚠️  Requires your review")
                
        blockers = path.get_blockers()
        if blockers:
            print(f"Blockers: {', '.join(blockers)}")

Correcting the Personal Model

修正Personal Model

python

from elephant_agent import PersonalModel

model = PersonalModel.load()

python

from elephant_agent import PersonalModel

model = PersonalModel.load()

Correct a misunderstanding

correction = model.correct( lens="identity", field="values", current="achievement, efficiency", corrected="learning, collaboration, impact", reason="I value learning and collaboration over pure efficiency" )

The model learns from corrections

print(f"Correction applied: {correction.applied}") print(f"Impact: {correction.impact_summary}")

undefined

print(f"Correction applied: {correction.applied}") print(f"Impact: {correction.impact_summary}")

undefined

Integrating with Workflows

工作流集成

python

from elephant_agent import ElephantAgent
import json

agent = ElephantAgent.load()

python

from elephant_agent import ElephantAgent
import json

agent = ElephantAgent.load()

Export context for external tools

def export_context_for_tool(tool_name: str): model = agent.get_personal_model()

context = {
    "current_focus": model.pulse.focus,
    "active_projects": model.world.projects,
    "constraints": model.pulse.constraints,
    "tool": tool_name
}

return json.dumps(context, indent=2)

def export_context_for_tool(tool_name: str): model = agent.get_personal_model()

context = {
    "current_focus": model.pulse.focus,
    "active_projects": model.world.projects,
    "constraints": model.pulse.constraints,
    "tool": tool_name
}

return json.dumps(context, indent=2)

Import learnings back

def import_learning(source: str, content: dict): model = agent.get_personal_model()

model.add_journey_entry({
    "source": source,
    "lesson": content["lesson"],
    "context": content["context"],
    "timestamp": content["timestamp"]
})

undefined

def import_learning(source: str, content: dict): model = agent.get_personal_model()

model.add_journey_entry({
    "source": source,
    "lesson": content["lesson"],
    "context": content["context"],
    "timestamp": content["timestamp"]
})

undefined

Troubleshooting

故障排查

Provider Connection Issues

服务商连接问题

bash

undefined

bash

undefined

Test provider connectivity

elephant config test-provider

Reset provider configuration

elephant config reset-provider

Check API key

echo $OPENAI_API_KEY # Should not be empty

Verify model availability

elephant models list

undefined

elephant models list

undefined

Personal Model Not Loading

Personal Model无法加载

python

from elephant_agent import PersonalModel

python

from elephant_agent import PersonalModel

Reset model if corrupted

PersonalModel.reset()

Reinitialize

agent.init(force=True)

Verify model structure

model = PersonalModel.load() assert model.is_valid(), "Model structure invalid"

undefined

model = PersonalModel.load() assert model.is_valid(), "Model structure invalid"

undefined

Dashboard Connection Issues

控制面板连接问题

bash

undefined

bash

undefined

Check if dashboard is running

elephant dashboard status

Kill existing dashboard

elephant dashboard stop

Restart on different port

elephant dashboard --port 8080

For remote access, use tunnel

ssh -L 3000:localhost:3000 user@remote-host

undefined

ssh -L 3000:localhost:3000 user@remote-host

undefined

Performance Optimization

性能优化

python

from elephant_agent import ElephantAgent

python

from elephant_agent import ElephantAgent

Reduce model calls with caching

agent = ElephantAgent( cache_responses=True, cache_ttl=300 # 5 minutes )

Lower curiosity effort for faster responses

agent.set_curiosity_effort("low")

Use smaller model for routine tasks

agent.set_model("gpt-3.5-turbo") # Faster, cheaper

Batch questions

questions = [ "What's my next step?", "Any blockers?", "Energy check?" ] responses = agent.ask_batch(questions)

undefined

questions = [ "What's my next step?", "Any blockers?", "Energy check?" ] responses = agent.ask_batch(questions)

undefined

Debugging

调试

bash

undefined

bash

undefined

Enable debug logging

export ELEPHANT_LOG_LEVEL=debug elephant wake

View logs

tail -f ~/.elephant/logs/elephant.log

Check runtime state

elephant debug state

Verify skills

elephant skills test filesystem

undefined

elephant skills test filesystem

undefined

Resources

资源

Documentation: https://elephant.agentic-in.ai/docs/
Paper: https://elephant.agentic-in.ai/paper/
Blog: https://elephant.agentic-in.ai/blog/personal-ai-you-create/
GitHub: https://github.com/agentic-in/elephant-agent
Homepage: https://elephant.agentic-in.ai/

文档: https://elephant.agentic-in.ai/docs/
论文: https://elephant.agentic-in.ai/paper/
博客: https://elephant.agentic-in.ai/blog/personal-ai-you-create/
GitHub: https://github.com/agentic-in/elephant-agent
官网: https://elephant.agentic-in.ai/

elephant-agent-personal-ai

Original

Translation

Elephant Agent Personal AI Skill

Elephant Agent 个人AI技能

What Elephant Agent Does

Elephant Agent 功能介绍

Installation

安装步骤

macOS Desktop App (Recommended)

macOS桌面应用（推荐）

Download from GitHub releases

Download from GitHub releases

Or visit releases page

Or visit releases page

CLI + Dashboard (Linux/Cloud/SSH)

CLI + 控制面板（Linux/云服务器/SSH）

Install via install script

Install via install script

Or manual installation

Or manual installation

Key CLI Commands

核心CLI命令

Initialization and Setup

初始化与配置

Initialize Elephant Agent (first run)

Initialize Elephant Agent (first run)

Check system readiness

Check system readiness

Configure provider settings

Configure provider settings

List all configuration

List all configuration

Daily Interaction

日常交互

Enter the chat TUI (main interaction mode)

Enter the chat TUI (main interaction mode)

Quick query without entering TUI

Quick query without entering TUI

View Personal Model

View Personal Model

View specific model lens

View specific model lens

Paths Management

Paths管理

List all paths

List all paths

Create a new path

Create a new path

View path details

View path details

Update path status

Update path status

Archive completed path

Archive completed path

Herd Management

Herd管理

List all agents in herd

List all agents in herd

Create a baby elephant for specific task

Create a baby elephant for specific task

View agent details

View agent details

Remove agent from herd

Remove agent from herd

Skills and Tools

技能与工具

List available skills

List available skills

Enable a skill

Enable a skill

Disable a skill

Disable a skill

View skill documentation

View skill documentation

Dashboard

控制面板

Open dashboard in browser

Open dashboard in browser

Run dashboard without opening browser (for remote/SSH)