Skip to main content
connic

Agents that remember and keep records.
Without a database to run.

Conversations pick up where they left off. Agents look up and update bookings or repairs and keep the files they produce. Every environment comes with its own storage, with nothing to set up.

9:41
OT
Old Town Hotel
Guest messages · HB-20417

Yes, the night desk is open until 02:00. I've noted your 23:30 arrival on the booking.

Today 08:15

Late checkout on Sunday?

Late checkout until 13:00 is free with your room. I've added it to your booking, next to the 23:30 arrival you mentioned yesterday.

Your booking HB-20417
Fri 6 Nov to Sun 8 Nov · Double room
Arrival around 23:30
Late checkout 13:00
Message Old Town Hotel

From group bookings to supplier invoices.

A travel desk, a hotel and a property office let agents look up, write and file their own data. Staff see the same records in their own apps.

An event planner asks for 40 rooms near the congress center. The agent finds the closest hotels with free rooms and drafts a proposal for the travel desk to check.

Group Travel
groups.example.net/requests/GR-5531
Group TravelRequestsHotelsInvoicesHannah Becker
GR-5531
Partner summit, 40 rooms
Nov 18 to 20Congress Center, Hall B
Proposal draft
HotelDistanceFree roomsPer night
Hotel am Park0.4 km24€142
Canal Suites0.9 km16€129
Fair Inn1.3 km40€118

24 and 16 rooms in the two closest hotels cover the group within 1 km. Fair Inn has all 40 in one place, 1.3 km away.

Received
Oct 30, 09:14 via request form
Drafted
Oct 30, 09:15
Status
Waiting for review
Send to clientEdit draft

Guests ask the hotel chat for pillows or a late checkout. The agent saves each request, and housekeeping marks it done on their board.

Housekeeping
staff.example.com/housekeeping/floor-4
Old Town HotelHousekeeping · Floor 4Ines Carvalho
Open requests4 open
  • 412
    2 extra pillows
    Guest chat
    6 minDone
  • 405
    Iron and ironing board
    Guest chat
    14 minDone
  • 407
    Fresh towels
    Front desk
    32 minDone
  • 418
    Late checkout 13:00 on Sunday, clean last
    Guest chat
    SunDone
Done today
  • 409Baby cot07:48
  • 403Extra blanket07:21

Once a month, an agent downloads supplier invoices and files them per building. The property office opens them in its own app, with amount and status.

Property Office
office.example.org/files/invoices/LIN-12/2026-10
Property OfficeBuildingsRepairsFilesJana Wolf
invoices
  • LIN-12
  • 2026-09
  • 2026-10
  • PARK-3
  • MILL-7
invoices / LIN-12 Lindenstrasse 12 / 2026-10
FileSupplierAmountStatus
water-2026-10.pdf
Oct 29, 06:12
City Water€412.80checked
power-common-areas-2026-10.pdf
Oct 29, 06:14
Grid Power€187.35new
elevator-service-q4.pdf
Oct 29, 06:15
Lift Service€1,140.00new
waste-2026-10.pdf
Oct 28, 06:10
Waste Services€96.40checked
4 files · saved by invoice-collector

Conversations continue the next day.

A tenant reports a repair in the evening and picks a slot the next morning. The agent still knows the flat, the repair and the slots it offered, because connic keeps each conversation's history between runs.

9:41
RS
Residents Service
Text message
Tuesday

Hi, the radiator in our living room has been knocking loudly since yesterday. Flat 3B, Lindenstrasse 12.

18:05

Thanks, I've logged repair R-1182 for flat 3B. The closest heating technician can come Thursday 8:00 to 10:00 or Friday 13:00 to 15:00. Which suits you?

18:05
Wednesday

Friday works. Should I bleed the radiator before?

07:52

Booked: Friday 13:00 to 15:00 for R-1182. No need to bleed it yourself, the technician checks the pressure first.

07:52
Session
Agentrepairs-assistant
Runs2
Last messageWed 07:52
Expires after 14 quiet days

See every record your agents keep.

Agents write repairs, bookings and requests into their environment's database. Your team browses, filters and corrects them in the dashboard, with no database server to run.

Search by value and distance
Filter by exact values, ranges or distance from a place, or describe the filter in plain words.
Start with the records you have
Import a file of up to 25 MB into an empty collection, such as the buildings and technicians an agent should know.
Files next to the records
Agents save reports and downloads into folders. The dashboard shows them, and each run links to the files it wrote.
Database471 docs
Collections
  • repairs412
  • technicians38
  • buildings21
repairs412 docsInsertSchema
{"status": "booked", "slot": {"$gte": "2026-11-02"}}
repair_idflattradeslot
R-1182LIN-12 · 3Bheating2026-11-06T13:00
R-1179LIN-12 · 5Celectrical2026-11-04T08:00
R-1177PARK-3 · 1Aplumbing2026-11-03T10:30
R-1174MILL-7 · 2Bheating2026-11-03T08:00

Decide in code what each agent may touch.

repairs-assistant keeps every text conversation, works with three collections and finds the closest technician with $near. The limits in its YAML apply on every run, also to your own tools.

session
One history per text conversation, keyed by conversation_id. ttl ends it after 14 days without a message.
concurrency
Runs for the same conversation wait for each other, so two quick messages don't collide.
database
Limits the agent to repairs, technicians and buildings, and prevent_delete blocks db_delete.
agents/repairs-assistant.yaml
version: "1.0"

name: repairs-assistant
description: Logs tenants' repair requests and books a technician
model: connic/gpt-5.6-terra
system_prompt: |
  Log repair requests from tenants and offer the
  closest technician's free slots. Use what the
  tenant already told you in this conversation.
tools:
  - repairs.nearest_technicians
  - db_insert
  - db_update
session:
  key: input.conversation_id
  ttl: 1209600
concurrency:
  key: conversation_id
  on_conflict: queue
database:
  collections: [repairs, technicians, buildings]
  prevent_delete: true
tools/repairs.py
from connic.tools import db_find


async def nearest_technicians(building: str, trade: str) -> list:
    """Find the three closest technicians for a trade, within 15 km.

    Args:
        building: Building code, e.g. LIN-12
        trade: heating, plumbing or electrical
    """
    found = await db_find("buildings", filter={"code": building})
    site = found["documents"][0]["location"]
    result = await db_find(
        "technicians",
        filter={
            "trade": trade,
            "location": {"$near": {**site, "maxDistance": 15000}},
        },
        limit=3,
    )
    # Nearest first, each with _distance in meters
    return result["documents"]

Frequently Asked Questions

No. Every environment comes with a database, conversation memory, file storage and a search index for documents. Collections and folders appear as soon as an agent writes to them, with no schema to design.

Yes. You name the collections, folders and document areas each agent may use, and you can block changes or deletions for all of them or one at a time. The limits also apply to your own tools.

No. Conversations, records, files and documents belong to one environment, so tests in staging never touch production data. You can copy selected collections across. Files are encrypted at rest, and each project keeps its platform data in its EU region.

There is nothing to provision. Stored data is billed from Project credit per GB-hour, files at their own rate plus the bytes downloaded. Files an agent reads through its file tools don't count as downloads. The pricing page lists the rates.