v3 · Live on Fly.io

Your AI operations team,
in one Telegram bot.

8 agents handle email triage, content drafting, research scanning, and community management. Policy-gated. Cost-tracked. Human-in-the-loop. Running 24/7 for under $10/month.

See it work → View source

What it does

One bot. Five commands. Zero daily effort.

Message @aigovops_bot on Telegram. It classifies your email, drafts your content, scans for research, tracks costs, and asks for approval before taking action. Everything is logged to an audit trail.

📧 Email Triage

Polls Gmail every 5 minutes. Classifies into 5 categories. Notifies you only when action is needed. Auto-archives the rest.

✍️ Content Drafting

draft about [topic] → AI writes a LinkedIn post in Foundation voice. Approve, reject, or edit with one tap.

🔬 Research Scanning

/research → scans for AI governance news, scores relevance, delivers a 3-item digest.

💰 Cost Tracking

Every LLM call is tracked per-agent. /costs shows a 7-day report. No surprise bills.

🛡️ Policy Engine

Cedar-style rules control what agents can do. Moderator can flag but never delete. Agents can't self-approve.

📋 Audit Trail

Every action logged with timestamp, agent, model, and outcome. /audit shows the last 10.

Architecture

Simple by design.

One Python process. SQLite for state. No Redis, no Kubernetes, no microservices.

Bob/Ken (Telegram)
    ↕
┌─────────────────────────────────────┐
│  run_bot.py (Fly.io, 24/7)         │
│                                     │
│  ├── Telegram handlers              │
│  ├── Email poller (IMAP, 5 min)    │
│  ├── PolicyEngine (Cedar rules)     │
│  ├── CostTracker (per-agent)        │
│  ├── StateStore (SQLite)            │
│  └── Audit log (append-only)        │
└─────────────────────────────────────┘
    ↕
┌─────────────────────────────────────┐
│  External APIs                      │
│  OpenAI · Gmail · Perplexity        │
└─────────────────────────────────────┘

Agents

YAML-defined. Policy-gated. Human-approved.

Each agent is a YAML file. No per-agent Python code. The engine interprets them all.

Agent	Trigger	What it does
Email Classifier	email.arrived	Classifies into 5 categories, drafts replies for action-required
Writing Agent	draft command	Writes LinkedIn posts in Foundation voice
Research Scanner	daily / on-demand	Finds AI governance news, scores relevance
Welcomer	member.joined	Sends personalized DMs to new community members
Moderator	post.published	Flags off-topic or spam content (can never delete)

Get started

Clone, configure, deploy.

git clone https://github.com/bobrapp/ai-bob-setup-agent.git
cd ai-bob-setup-agent
cp .env.example .env          # Add your API keys
python3 -m pytest tests/ -v   # 83 tests pass
fly deploy --ha=false --yes   # Ship it