Skip to content

19-use-cases.md

Latest commit

 

History

History
391 lines (298 loc) · 9.4 KB

19-use-cases.md

File metadata and controls

391 lines (298 loc) · 9.4 KB
Raw

Automation Use Cases

This guide maps seven real-world automation patterns to sven's configuration and tools. Each use case combines multiple integrations into a cohesive workflow.

1. Autonomous Personal CRM

Analyze emails and calendar events to build a knowledge graph of contacts, track action items, and prepare context before calls.

What the agent does

  • Reads incoming emails and extracts contact details, preferences, and action items
  • Saves everything to semantic memory with entity tags
  • Reviews context before meetings (pulling related memories)
  • Sends follow-up reminders via Telegram

Configuration

# ~/.config/sven/node.yaml
scheduler:
  heartbeat:
    enabled: true
    every: "1h"

channels:
  telegram:
    bot_token: "${TELEGRAM_BOT_TOKEN}"
    allowed_users: [123456789]
# ~/.config/sven/config.yaml
tools:
  email:
    backend: imap
    imap_host: "imap.gmail.com"
    username: "${EMAIL_USER}"
    password: "${EMAIL_PASSWORD}"
  memory:
    backend: sqlite
  calendar:
    backend: caldav
    url: "${CALDAV_URL}"
    username: "${CALDAV_USER}"
    password: "${CALDAV_PASSWORD}"

HEARTBEAT.md

# CRM Heartbeat

Every hour:
1. List unread emails (email list unread_only=true limit=20)
2. For each email, extract: sender name, company, phone, key facts, action items
3. Save extracted info with semantic_memory remember (entity=name, source=email, tags=[contact,crm])
4. Check tomorrow's calendar events; for each meeting attendee, recall their profile
5. If any action item is overdue, send reminder via Telegram

2. Business on Autopilot

Monitor competitors, scrape pricing, draft content, manage email, and track campaign metrics.

Configuration

scheduler:
  heartbeat:
    enabled: true
    every: "2h"
  jobs:
    # defined via schedule tool at runtime
tools:
  email:
    backend: gmail
    oauth_client_id: "${GMAIL_CLIENT_ID}"
    oauth_client_secret: "${GMAIL_CLIENT_SECRET}"

Example Scheduled Jobs

Create these once via the schedule tool:

schedule create "competitor-monitor" with cron "0 */4 * * *" and prompt
"Fetch https://competitor.example.com/pricing. Compare with last week's prices saved in memory.
If any price changed by more than 5%, send a Telegram alert."

schedule create "content-draft" with cron "0 9 * * 1" and prompt
"Draft 3 LinkedIn posts about recent industry news. Save drafts to /workspace/content/linkedin-drafts.md"

schedule create "email-triage" with cron "0 8,13,18 * * *" and prompt
"List unread emails. Auto-reply to common inquiries using templates in /workspace/email-templates/.
Flag anything requiring personal attention."

3. Proactive Daily Briefings

Wake up, review tasks, emails, news, and send a personalized summary via Telegram or WhatsApp.

Configuration

scheduler:
  heartbeat:
    enabled: false  # Use cron instead for precise timing

channels:
  telegram:
    bot_token: "${TELEGRAM_BOT_TOKEN}"
    allowed_users: [123456789]

Scheduled Job

{
  "action": "create",
  "name": "morning-briefing",
  "cron": "0 7 * * *",
  "prompt": "Morning briefing:\n1. Fetch top 5 headlines from https://news.ycombinator.com\n2. List today's calendar events\n3. List urgent unread emails\n4. Check open tasks\n5. Compose a concise summary and send to Telegram chat 123456789",
  "deliver_to": "telegram:123456789"
}

Output Example

The agent sends you:

Good morning! Here's your briefing for Tuesday, April 15:

📅 Today's Schedule:
• 9:00 AM — Standup with engineering team
• 2:00 PM — Strategy call with Alice (Acme Corp)
• 4:30 PM — 1:1 with Bob

📧 Urgent Emails (3):
• Invoice from Cloudflare — due Friday
• Alice: "Can we move the call to 3pm?" → Replied ✓
• New customer inquiry from TechCorp

📰 Top News:
• OpenAI releases GPT-5 with 1M context
• Rust 2.0 stabilizes async traits
• Y Combinator opens W2026 applications

✅ Open Tasks: 4 (2 due today)

4. 3D Model Generation and Search

Research existing 3D models online or generate custom ones using AI tools.

Configuration

No special integrations required — uses web search and shell tools.

Example Prompts

Search Thingiverse for "Raspberry Pi 5 case with fan mount". Download the top 3 results,
evaluate their dimensions, and recommend the best fit for a 40mm fan.

I need a custom bracket for a 28mm diameter pipe with a 15-degree angle.
Research if any existing models match. If not, generate OpenSCAD code for it and save to /workspace/bracket.scad

Skill Template

Create /workspace/.sven/skills/3d-models.md:

# 3D Model Research and Generation

## Search
- Thingiverse: https://www.thingiverse.com/search?q={query}
- Printables: https://www.printables.com/search/models?q={query}
- MyMiniFactory: https://www.myminifactory.com/search/?query={query}

## Generation
- Use OpenSCAD syntax for parametric models
- Default resolution: $fn=50
- Include comments explaining parameters
- Save to /workspace/models/{name}.scad

5. Second Brain Knowledge Management

Text anything to remember — the agent builds a searchable local knowledge base.

Configuration

channels:
  telegram:
    bot_token: "${TELEGRAM_BOT_TOKEN}"
    allowed_users: [123456789]

tools:
  memory:
    backend: sqlite

HEARTBEAT.md

# Second Brain Agent

When a user sends a message:
- If it looks like something to remember (fact, idea, contact, task): save with semantic_memory remember
- If it's a question: search memory with semantic_memory recall and answer
- Categorize automatically: people, ideas, tasks, references, notes
- Confirm what was saved in your reply

Usage

Just text your Telegram bot:

  • "Remember: dentist appointment April 20 at 2pm"
  • "Alice from Acme prefers calls after 2pm, doesn't do Mondays"
  • "Good article on Rust async: https://..."
  • "What's Alice's preference for meetings?"
  • "What do I know about Acme Corp?"

6. Automated Content Creation

Watch batched videos, write captions, and schedule uploads.

Configuration

scheduler:
  heartbeat:
    enabled: true
    every: "6h"

Workflow

  1. Drop video files in /workspace/content/raw/
  2. Agent detects new files and processes them:
# Content Creation Heartbeat

Check /workspace/content/raw/ for new video files.
For each unprocessed video:
1. Get video metadata with: run_terminal_command ffprobe -v quiet -print_format json -show_format {file}
2. If it's a short (< 90 seconds), generate a TikTok/Reels caption
3. If it's longer, generate a YouTube title, description, and tags
4. Save metadata to /workspace/content/metadata/{filename}.json
5. Move processed files to /workspace/content/ready/

Skill Template

# Video Content Guidelines

## Short-form (TikTok/Reels)
- Hook in first 3 words
- 3-5 hashtags
- Call to action at end
- Max 150 chars

## YouTube
- Title: benefit-driven, 60 chars max, include keyword
- Description: first 150 chars is above fold
- Include timestamps for videos > 10 min
- Tags: 10-15, mix broad and specific

7. AI Voice Call Agents

Make voice calls to confirm appointments, collect information, and summarize conversations.

Configuration

tools:
  voice:
    tts_provider: "elevenlabs"
    tts_api_key: "${ELEVENLABS_API_KEY}"
    tts_voice_id: "21m00Tcm4TlvDq8ikWAM"
    call_provider: "twilio"
    twilio_account_sid: "${TWILIO_SID}"
    twilio_auth_token: "${TWILIO_TOKEN}"
    twilio_phone_number: "+12065551234"

  calendar:
    backend: caldav
    url: "${CALDAV_URL}"

Appointment Confirmation

Create a scheduled job that calls clients before appointments:

{
  "action": "create",
  "name": "appointment-confirmations",
  "cron": "0 16 * * *",
  "prompt": "Check tomorrow's calendar events that include an attendee phone number in the description. For each unconfirmed appointment, place a confirmation call. Use the voice call tool with a friendly script: 'Hi, this is sven calling on behalf of [your name]. I'm calling to confirm your appointment tomorrow at [time]. Please call back to reschedule if needed. Thank you!'. Save call result to memory."
}

Usage

Make a call to +12065551234 to confirm the 2pm appointment tomorrow.
Tell them it's with Alice at Acme Corp and they can reschedule by calling back.

Transcribe the recording at /recordings/client-call-2026-04-15.mp3
and save the key points and any action items to memory.

Complete Example Config

# ~/.config/sven/node.yaml
scheduler:
  heartbeat:
    enabled: true
    every: "30m"
    prompt: "Heartbeat: check inbox, calendar, pending tasks. Act on urgent items."

channels:
  telegram:
    bot_token: "${TELEGRAM_BOT_TOKEN}"
    allowed_users: [123456789]

hooks:
  token: "${HOOKS_TOKEN}"
  mappings:
    gmail:
      path: "/hooks/gmail"
      prompt: "New email notification. Process inbox."

# ~/.config/sven/config.yaml
tools:
  email:
    backend: imap
    imap_host: "imap.gmail.com"
    username: "${EMAIL_USER}"
    password: "${EMAIL_PASSWORD}"
  calendar:
    backend: caldav
    url: "${CALDAV_URL}"
    username: "${CALDAV_USER}"
    password: "${CALDAV_PASSWORD}"
  memory:
    backend: sqlite
  voice:
    tts_provider: "elevenlabs"
    tts_api_key: "${ELEVENLABS_API_KEY}"
    call_provider: "twilio"
    twilio_account_sid: "${TWILIO_SID}"
    twilio_auth_token: "${TWILIO_TOKEN}"
    twilio_phone_number: "+12065551234"