Skip to main content
This guide walks you through setting up Virtual Persona from scratch — configuring your identity, generating your persona, enabling monitoring, and gradually granting auto-respond access.

​
Prerequisites

Teams MCP server connected and authenticated (msteams in /mcp)
Telegram transport configured (for receiving private VP notifications)
At least 20+ recent Teams conversations (for quality persona generation)

​
Step 1: Configure Your Identity

Add your email address(es) to .env so VP knows which messages are yours: 
VP_SELF_EMAILS=you@company.com
Multiple emails? Comma-separate them: 
VP_SELF_EMAILS=you@company.com,you@personal.com
This is required. Without VP_SELF_EMAILS, VP can’t distinguish your messages from incoming ones and will try to process your own messages.

​
Step 2: Enable VP

Start Octo and run: 
/vp enable
This scaffolds all data files in .octo/virtual-persona/ and starts the polling loop. Check that everything is running: 
/vp status
You should see:
  • VP enabled: yes
  • Poller running: yes
  • Allow list: 0 users (this is correct — monitor-only mode)

​
Step 3: Generate Your Persona

This is the most important step. VP needs to learn your communication style from your actual conversations. 
/vp persona generate Your Name
The pipeline takes several minutes and runs through five phases:
  1. Fetching conversations — pulls 30-50 Teams chats via MCP
  2. Per-conversation analysis — analyzes each chat individually (you’ll see progress)
  3. Web research — searches for your public presence (articles, talks, GitHub)
  4. Synthesis — merges everything into a communication profile
  5. Prompt generation — creates the system prompt that defines your VP voice
Persona generation uses Octo’s full agent stack for web research, synthesis, and prompt generation. This gives much richer results than simple LLM calls.

​
Review the Results

/vp persona show       # Preview the generated prompt
/vp persona analysis   # See your communication profile summary
The generated files are human-editable:
  • Prompt: .octo/virtual-persona/system-prompt.md — edit to fine-tune your voice
  • Analysis: .octo/virtual-persona/communication-analysis.json — your behavioral profile
  • Web research: .octo/virtual-persona/web-research.md — public presence findings
Re-run /vp persona generate any time. Each run replaces the previous persona. You can also manually edit system-prompt.md for fine-tuning — VP loads it fresh on each message.

​
Step 4: Monitor Mode (Passive)

At this point VP is running but in monitor-only mode. For every incoming Teams message:
  1. VP reads and classifies it
  2. If it needs a response, VP generates a suggested answer via Octo
  3. The suggestion is sent to you privately (Telegram/console) — never to the sender
  4. You decide whether to reply yourself or ignore it
Check what VP is seeing: 
/vp audit 5            # Last 5 decisions
/vp stats 1            # Today's stats
/vp threads 5          # Classified thread topics
Monitor mode is the safest way to start. You see every VP decision without any risk of unwanted messages being sent. Stay in this mode until you’re comfortable with VP’s classification accuracy.

​
Step 5: Allow Auto-Responses

Once you trust VP’s judgment for specific people, grant them auto-respond access: 
/vp allow alice@company.com
Now when Alice messages you, VP will auto-respond if confidence is ≥80%.

​
Confidence Tuning

Adjust per-user confidence with modifiers: 
/vp allow alice@company.com +10     # Alice gets a confidence boost
/vp allow bob@company.com -15       # Be more conservative with Bob

​
Dry-Run Testing

Test VP’s decision for a specific person without sending anything: 
/vp test alice@company.com "How do I set up LangGraph agents?"
This shows: access decision, confidence score, category, and reasoning.

​
Step 6: Handle Escalations

When VP can’t handle a message confidently, it escalates silently:
  • You receive a private Telegram notification with full context, confidence score, and a suggested answer
  • The thread is locked — VP won’t respond until you release it
  • The sender sees nothing — they’re just waiting for your reply

​
Responding via Telegram

The fastest way to handle escalations is to reply directly to the Telegram notification:
  • Reply with text — your reply is sent to the Teams chat, delegation auto-released
  • Reply “ignore” — the chat is muted, delegation released
No need to open Teams or run slash commands — just reply naturally.

​
Manual Release

If you handled it in Teams directly: 
/vp release <chat_id>     # Release a specific thread
/vp release all            # Release all locked threads
View all locked threads: 
/vp delegated

​
Step 7: Priority Users & Ignoring Chats

​
Priority Users

Some contacts should never be ignored — their messages should always be processed even in muted chats: 
/vp priority add boss@company.com
/vp priority                        # List priority users

​
Ignoring Noisy Chats

Mute bot channels, automated notifications, or large group chats: 
/vp ignore <chat_id> "Bot notifications"
/vp ignored                         # List ignored chats
/vp unignore <chat_id>              # Unmute
Run /vp sync first to see chat IDs in the logs, then copy the ID of the chat you want to ignore.

​
Step 8: People Profiles

VP tracks interaction patterns. View someone’s profile: 
/vp profile alice@company.com
Set a tone override if VP’s default isn’t right for someone: 
/vp profile alice@company.com tone semi_formal
Available tones: casual_technical (default for engineers), casual, semi_formal

​
Configuration Reference

All VP settings via .env:
VariableDefaultDescription
VP_ENABLEDtrueMaster switch
VP_SELF_EMAILS(empty)Your email(s), comma-separated
VP_POLL_INTERVAL2mHow often to check Teams (30s, 2m, 1h, or bare seconds)
VP_ACTIVE_HOURS_START08:00Start of active hours (local time)
VP_ACTIVE_HOURS_END22:00End of active hours (local time)

​
Gradual Rollout Strategy

1

Week 1: Monitor Only

Enable VP with an empty allow list. Watch decisions via /vp audit and /vp stats. Tune the persona prompt if needed.
2

Week 2: Allow Close Collaborators

Add 2-3 people you work with daily. Monitor their auto-responses via /vp audit. Adjust confidence modifiers.
3

Week 3+: Expand Gradually

Add more people as you gain confidence in VP’s accuracy. Use /vp block for anyone who should always go to you directly.

​
Troubleshooting

Check that VP_SELF_EMAILS is set correctly in .env. Without it, VP may skip all messages or try to process your own.
Ensure the Teams MCP server is authenticated. Run /mcp to check server status. You may need to re-authenticate with mcp__msteams__login.
Re-run /vp persona generate with more conversations available, or manually edit .octo/virtual-persona/system-prompt.md to adjust phrasing and rules.
VP is being too conservative. Lower the confidence threshold by adding positive modifiers to your allowed users: /vp allow user@company.com +15
Add the user to the block list: /vp block user@company.com. Review the audit log: /vp audit 10. Consider lowering their confidence modifier.

VP Feature Reference

Full VP architecture and command reference

Proactive AI

Heartbeat and cron — Octo’s other proactive features