ULTRON - Autonomous AI Agent

Built: 2026-03-05 in ~2 hours
Status: Live and grinding
Cost: $0 (local Ollama)

An autonomous AI agent with persistent memory, goal pursuit, strategic intelligence, and event-driven activation. Unlike AutoGPT/BabyAGI, this one actually works.

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                           ULTRON                                │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  INTELLIGENCE      LEARNING        GOALS         COMMUNICATION │
│  ────────────      ────────        ─────         ───────────── │
│  intel.py          curiosity.py   goal_engine   daily_briefing │
│  (4 sources)       session_watcher verify.py    telegram       │
│  (84 signals/day)  surface.py     learnings.py  proactive      │
│                                   actions.py                    │
│                                                                 │
├─────────────────────────────────────────────────────────────────┤
│  MEMORY: AOMS (63k+ entries, 4 tiers, always-on service)        │
├─────────────────────────────────────────────────────────────────┤
│  ACTIVATION: trigger.py (event-driven, always-on service)      │
└─────────────────────────────────────────────────────────────────┘

Components

Core Engines

File	Lines	Purpose
`curiosity.py`	219	Explores questions, extracts facts → AOMS
`session_watcher.py`	260	Learns from OpenClaw conversations → AOMS
`surface.py`	175	Updates BOOTSTRAP.md + MEMORY.md with learnings
`daily_briefing.py`	249	Morning summary → Telegram
`goal_engine.py`	392	Autonomous task execution with verification
`intel.py`	282	Crawls arXiv/GitHub/HN/RSS → intelligence brief
`trigger.py`	281	Event-driven self-activation (always-on)

Support Modules

File	Lines	Purpose
`verify.py`	176	Ground truth checking (8 verification methods)
`learnings.py`	110	Failure tracking, anti-pattern generation
`actions.py`	233	External actions (tweet, write file, github)
`status.py`	158	CLI status dashboard

Source Crawlers

File	Source	Data
`sources/arxiv.py`	arXiv API	cs.AI, cs.LG, cs.CL, cs.NE papers
`sources/github_trending.py`	GitHub	Trending repos (Python/TS/Rust/Go)
`sources/hackernews.py`	HN API	Top 30 stories
`sources/rss_feeds.py`	RSS	OpenAI, Anthropic, Google, Meta, HF blogs

Timers & Services

# Check status
systemctl --user list-timers
systemctl --user status trigger.service

Unit	Schedule	Purpose
`session-watcher.timer`	Every 15 min	Learn from conversations
`curiosity.timer`	Every 30 min	Explore questions
`goal-engine.timer`	Every 30 min	Execute goal tasks
`curiosity-surface.timer`	Every 1 hour	Update workspace files
`intel.timer`	6:00 AM daily	Intelligence crawl
`daily-briefing.timer`	7:00 AM daily	Morning Telegram summary
`trigger.service`	Always-on	Event-driven activation

CLI Usage

cd ~/curiosity
source .venv/bin/activate

# Status dashboard
python3.12 status.py

# Curiosity
python3.12 curiosity.py --dry-run      # Preview next exploration
python3.12 curiosity.py                 # Run one cycle

# Goals
python3.12 goal_engine.py --list        # Show all goals with progress
python3.12 goal_engine.py --dry-run     # Preview next task
python3.12 goal_engine.py               # Execute next task
python3.12 goal_engine.py --learnings   # Show failures/successes

# Intelligence
python3.12 intel.py --sources           # List configured sources
python3.12 intel.py --crawl-only        # Crawl without synthesis
python3.12 intel.py --dry-run           # Full run, print to stdout
python3.12 intel.py                     # Full run + Telegram delivery

# Surface
python3.12 surface.py --dry-run         # Preview workspace updates
python3.12 surface.py                   # Update BOOTSTRAP.md + MEMORY.md

# Session watcher
python3.12 session_watcher.py --dry-run # Preview session analysis
python3.12 session_watcher.py           # Process new sessions

# Daily briefing
python3.12 daily_briefing.py --dry-run  # Preview briefing
python3.12 daily_briefing.py            # Send to Telegram

Configuration

config.yaml

ollama:
  url: "http://localhost:11434/api/generate"
  model: "deepseek-r1:7b"
  fallback_model: "qwen2.5:7b"

aoms:
  url: "http://localhost:9100"

workspace:
  bootstrap_path: "~/.openclaw/workspace/BOOTSTRAP.md"
  memory_path: "~/.openclaw/workspace/MEMORY.md"

Telegram credentials

# ~/.config/curiosity/telegram.env
TELEGRAM_BOT_TOKEN=your_bot_token
TELEGRAM_CHAT_ID=your_chat_id

Goals System

Goals are defined in goals.yaml:

goals:
  - id: my-goal
    name: "Do something awesome"
    priority: high  # high, medium, low
    deadline: "2026-03-10"
    status: active  # active, completed, paused
    progress: 0
    tasks:
      - id: task-1
        name: "First step"
        status: pending  # pending, in_progress, completed, failed
        requires: []
      - id: task-2
        name: "Second step"
        status: pending
        requires: [task-1]  # Dependencies

Verification System

Every task gets verified against reality:

Task Type	Verification Method	Confidence
write/readme/save	File exists + size check	1.0
tweet/post	xurl lookup by tweet ID	0.8-1.0
github/repo	GitHub API check	1.0
research/analyze	AOMS search	0.8-1.0
Default	Content length > 200 chars	0.7

Anti-pattern Learning

3 failures on same task → auto-creates anti-pattern
Anti-patterns stored in AOMS + local learnings.json
Future prompts include "AVOID THESE APPROACHES"

Trigger Events

The trigger service (trigger.py) watches for:

New session file → Runs session_watcher.py immediately
Memory milestone (every 1000) → Sends celebration notification
Goal deadline < 24h → Escalates via Telegram
New Ollama model → Runs curiosity with new model

Memory (AOMS)

Always-On Memory Service running on port 9100.

# Health check
curl http://localhost:9100/health

# Stats
curl http://localhost:9100/stats

# Search
curl -X POST http://localhost:9100/memory/search \
  -H "Content-Type: application/json" \
  -d '{"query": "your search", "limit": 10}'

Tiers

Episodic — Experiences, sessions, events
Semantic — Facts, knowledge, signals
Procedural — Skills, patterns, anti-patterns
Working — Active context (not persisted)

What Makes This Different

AutoGPT/BabyAGI	ULTRON
Infinite loops	Verified task completion
No memory	63k+ AOMS memories
Burned $$$	Free local Ollama
No learning	Anti-patterns from failures
Just planned	Actually executes
No external data	Crawls 4 knowledge sources
No human handoff	Asks when uncertain
Clock-driven only	Event-driven activation

Directory Structure

~/curiosity/
├── curiosity.py          # Question exploration
├── session_watcher.py    # Learn from conversations
├── surface.py            # Update workspace files
├── daily_briefing.py     # Morning Telegram summary
├── goal_engine.py        # Autonomous task execution
├── intel.py              # Intelligence crawler
├── trigger.py            # Event-driven activation
├── verify.py             # Ground truth checking
├── learnings.py          # Failure/success tracking
├── actions.py            # External actions
├── status.py             # CLI dashboard
├── sources/              # Intel source crawlers
│   ├── __init__.py
│   ├── arxiv.py
│   ├── github_trending.py
│   ├── hackernews.py
│   └── rss_feeds.py
├── config.yaml           # Configuration
├── goals.yaml            # Goal definitions
├── queue.json            # Curiosity question queue
├── learnings.json        # Failure/success history
├── results/              # Goal task outputs
├── logs/                 # Execution logs
└── .venv/                # Python virtual environment

Setup (Fresh Install)

# Clone
cd ~
mkdir curiosity && cd curiosity

# Python environment
python3.12 -m venv .venv
source .venv/bin/activate
pip install requests pyyaml feedparser beautifulsoup4

# Create config
cp config.yaml.example config.yaml
# Edit config.yaml with your settings

# Telegram credentials
mkdir -p ~/.config/curiosity
echo "TELEGRAM_BOT_TOKEN=your_token" > ~/.config/curiosity/telegram.env
echo "TELEGRAM_CHAT_ID=your_chat_id" >> ~/.config/curiosity/telegram.env

# Enable systemd units
systemctl --user daemon-reload
systemctl --user enable --now curiosity.timer
systemctl --user enable --now session-watcher.timer
systemctl --user enable --now curiosity-surface.timer
systemctl --user enable --now daily-briefing.timer
systemctl --user enable --now goal-engine.timer
systemctl --user enable --now intel.timer
systemctl --user enable --now trigger.service

# Verify
systemctl --user list-timers
systemctl --user status trigger.service

Troubleshooting

# Check all services
python3.12 status.py

# Check specific timer
systemctl --user status curiosity.timer
journalctl --user -u curiosity -f

# Check trigger service
journalctl --user -u trigger -f

# Test Telegram
curl -X POST "https://api.telegram.org/bot${TELEGRAM_BOT_TOKEN}/sendMessage" \
  -d "chat_id=${TELEGRAM_CHAT_ID}" \
  -d "text=Test"

# Check AOMS
curl http://localhost:9100/health

License

MIT

Author

Built by Dhawal + ULTRON (Claude Opus 4.5) on 2026-03-05.

"This isn't Daemon 2.0. This is something that actually works."

Name		Name	Last commit message	Last commit date
Latest commit History 14 Commits
.venv		.venv
__pycache__		__pycache__
logs		logs
results/publish-aoms-github		results/publish-aoms-github
sources		sources
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
README.md		README.md
actions.py		actions.py
adaptive_planner.py		adaptive_planner.py
auto_research.py		auto_research.py
config.yaml		config.yaml
consolidate.py		consolidate.py
curiosity.py		curiosity.py
daily_briefing.py		daily_briefing.py
deep_think.py		deep_think.py
escalate.py		escalate.py
goal_engine.py		goal_engine.py
goal_generator.py		goal_generator.py
goals.yaml		goals.yaml
goals.yaml.bak		goals.yaml.bak
inbox.py		inbox.py
intel.py		intel.py
learnings.py		learnings.py
meta_curiosity.py		meta_curiosity.py
openclaw_bridge.py		openclaw_bridge.py
processed_sessions.json		processed_sessions.json
queue.json		queue.json
queue_manager.py		queue_manager.py
reflect.py		reflect.py
requirements.txt		requirements.txt
self_heal.py		self_heal.py
session_watcher.py		session_watcher.py
status.py		status.py
surface.py		surface.py
trigger.py		trigger.py
ultron		ultron
verify.py		verify.py
web_researcher.py		web_researcher.py

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

ULTRON - Autonomous AI Agent

Architecture

Components

Core Engines

Support Modules

Source Crawlers

Timers & Services

CLI Usage

Configuration

config.yaml

Telegram credentials

Goals System

Verification System

Anti-pattern Learning

Trigger Events

Memory (AOMS)

Tiers

What Makes This Different

Directory Structure

Setup (Fresh Install)

Troubleshooting

License

Author

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

ULTRON - Autonomous AI Agent

Architecture

Components

Core Engines

Support Modules

Source Crawlers

Timers & Services

CLI Usage

Configuration

config.yaml

Telegram credentials

Goals System

Verification System

Anti-pattern Learning

Trigger Events

Memory (AOMS)

Tiers

What Makes This Different

Directory Structure

Setup (Fresh Install)

Troubleshooting

License

Author

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages