Skip to main content

Pinche, your AI assistant

Available on the Advanced and Elite plans. Pinche is a conversational assistant integrated into RestaPro that understands natural language in Spanish and runs actions on your restaurant.

The name comes from pinche de cocina (the kitchen helper) - the assistant who knows everything.

How to access it​

Floating button (chef icon) in the bottom-right corner of any backoffice screen. Click → side chat panel.

  • Enter sends the message.
  • Esc closes the panel.
  • Conversations are stored per session (they don't persist across days for privacy).

25 tools (10 write + 15 read)​

Read - Pinche can look up everything​

Operations:

  • list_today_reservations - today's reservations with status filter
  • list_open_reservations_week - next 7 days
  • get_opening_hours - current schedule
  • list_service_pauses - active pauses
  • list_table_blocks - blocked tables

Menu:

  • get_menu_summary - categories + dishes + prices + visibility

Customers:

  • search_customer - search by name, phone, email
  • get_customer_history - full customer history
  • get_reviews_summary - Google rating + request funnel

Restaurant:

  • get_restaurant_stats - quick recap of the latest period
  • get_embed_code - iframe ready to copy
  • get_public_urls - menu, reservations, embed, QR URLs

Marketing:

  • list_marketing_campaigns - campaign history
  • list_automation_rules - active rules

Events:

  • list_group_event_proposals - proposals with status filter

Write - Pinche can modify (with confirmation)​

  • update_opening_hours - change hours
  • update_menu_item - price, name, description, visibility
  • update_reservation_status - confirm, cancel, no-show, etc.
  • toggle_bookings - turn reservations on/off globally
  • update_restaurant_info - phone, email, description
  • pause_service - one-off service pause
  • resume_service_pause - resume
  • toggle_automation_rule - activate/pause an email rule
  • respond_group_event_proposal - accept or decline a proposal
  • block_table - block a table for maintenance

Natural-language examples​

You: How many reservations do I have today?
Pinche: You have 14 confirmed reservations today. The first at 13:00 (Ana, 4 pax).

You: Close dinner reservations for Saturday
Pinche: I'm going to pause Saturday May 11 dinner reservations.
        Confirm?
You: Yes
Pinche: Done. Auto-resume when the day changes.

You: Bump the sirloin to 32 euros
Pinche: I found "Sirloin steak" on the menu (28€).
        I'll change it to 32€. Confirm?
You: Go ahead
Pinche: Updated.

You: Block table 4 tomorrow from 14 to 16 for maintenance
Pinche: I'll block Table 4 (Dining room) on May 5 from 14:00 to 16:00,
        reason "maintenance". Confirm?

Permissions by role​

  • Read: any role (owner, admin, staff).
  • Write: only owner and admin. If you're staff and ask for a change, Pinche replies "You don't have enough permissions to modify restaurant data. Ask an administrator to do it."

Automatic validations​

Before executing:

  • Existence: the dish/customer/reservation exists.
  • Ownership: the data belongs to your restaurant (not another in the workspace by mistake).
  • Valid ranges: HH:MM hours, future dates for pauses, ISO 8601 in blocks.
  • Explicit confirmation: never executes changes without your OK.

Context Pinche receives​

In each conversation, Pinche has access to:

  • Restaurant data (name, slug, timezone, address, phone).
  • The day's dashboard (reservations, KPIs).
  • Reservation configuration (mode, slot, duration, window).
  • Opening hours.
  • Tables and floor plan.
  • Menu summary (categories + items with prices).
  • Workspace (plan, members).
  • Active service pauses (so it doesn't recommend reserving in closed slots).

Logs and costs​

Every conversation is logged in pinche_chat_logs:

  • Who, when, message, response, tools used.
  • Tokens consumed, estimated cost in EUR.

Model: Google's Gemini 2.5 Flash. Fast (200-500 ms) and cheap.

Typical cost: under 0.02€ per conversation (input $0.30/1M tokens, output $2.50/1M, cached $0.075/1M with 75% discount).

These costs are included in your plan, no extra billing.

Privacy​

  • Logs are private to your workspace (RLS).
  • Gemini processes the messages but doesn't use them to train models (Vertex AI no-training).
  • Don't send sensitive data the model shouldn't see (Pinche is safe but the context goes to Google).

Best practices​

  • Be specific: "Today's reservations" > "Reservations".
  • Chain: after a reply, follow up with "And Saturday?". Pinche keeps context.
  • Confirm before changes: always read the summary before saying "yes".
  • For long tasks: use the backoffice. Pinche is for one-off actions and quick lookups.
Last updated on
ESENCA