French estate agency CRM and AI agents: which one to connect, and what it changes for an agency

This guide covers the connection between an agent and the agency software. The tasks an agent takes on once connected are covered in five AI automations that work in estate agencies, and connecting a CRM straight to Claude in our guide Claude and your CRM.

This matters if you want to connect an agent. Comparing Hektor with La Boîte Immo, or AC3 with Immofacile, means comparing the same publisher twice, and so the same data access policy. Here is what each official website said on 29 October 2026.

Orisha Real Estate traces AC3 on its history page: Poliris and Immofacile left the SeLoger group to join Gercop, "donnant naissance au Groupe AC3". For Netty the evidence is technical: netty.fr answers with a permanent redirect to modelo.fr, and the Modelo Office product page still uses images named "netty". The modelo.fr site is published by Septeo Real Estate Solutions (modelo.fr legal notice, read on 29 October 2026).

The documentation is the publisher's own. On apimo.com/api, the "Api3 Documentation" button leads to apimo.com/api/webservice, a page that loads the spec from the Apiwork/apimo-openapi GitHub repository; Apiwork SAS, registered in Cannes, publishes Apimo according to the apimo.com legal notice read on 29 October 2026. Apiwork rewrote the spec on the very day of our check.

We counted it on commit 3a8911f of 29 October 2026: 51 operations, with 29 reads, 12 creates, 7 updates and 3 deletes. The three deletes cover a listing's photos, media and documents. An agent therefore cannot delete a contact or a lead through the API, which limits the damage of a mistake. Authentication is HTTP Basic with the provider ID and a token, and each route requires a right (scope) enabled on the account.

Three practical limits. First, with no documented webhook, the agent learns about a change (a listing sold, a new lead entered by a negotiator) only on its next pass, and every call counts against a default quota of 1,000 requests a day, lead creation excepted. Second, contact creation does not check for duplicates: the spec states that each call creates a new contact, so the agent searches before it creates. Third, the API has no message-sending operation: email or SMS follow-ups go through a third-party tool, or through Automator, the built-in automation Apimo presented on 12 December 2024 with email, SMS and push reminders.

Third-party tools already connect: on 10 September 2026 Apimo announced its Leedflow integration, an assistant that summarises calls. On 29 October 2026, apimo.com shows 89 euros a month for CRM AI and 139 euros a month for CRM AI+; the page says neither whether that is per user nor whether VAT is included, and the cost of a partner account is not published.

Re-read on 4 November 2026 at commit 34b6c88 of 30 October 2026, the spec still has 51 operations and no webhook. The blocks below turn it into a working plan for an agency that wants to connect an agent: access, quota, what the agent writes, what it cannot do, how follow-ups start, and who signs off.

Getting access: one partner account, one connection per agency

The agent authenticates with the provider ID and token of a partner account, over HTTP Basic and HTTPS only. The spec sets three access rules. The account only sees agencies that have activated a connection with it. Every route needs a right (scope) enabled on the account, otherwise the API returns 403. An IP allow-list can be switched on for the account.

As of 4 November 2026, neither the Web service & API page nor Apimo's contact page explains how to obtain that account, how long it takes or what it costs. The request therefore goes to Apimo, from the provider building the agent. The agency then activates the connection on its side. A connection can cover one agency, a company, a network or a single agency sector.

The quota and timestamp sync

Every call uses one request from a daily quota, 1,000 by default, reset every night. Beyond that, the API returns 429. Two writes are not counted: sending a lead, and updating a listing's publication on a portal.

Lists are sorted by last update, oldest first. On each pass the agent sends back the timestamp it received last time and gets only what changed. A list that spans several pages costs one request per page. Polling three lists every five minutes already uses 864 requests a day (3 × 288), which leaves 136 for reading a record in detail or writing. So the pace is set list by list: leads often, listings and contracts a few times a day.

What the agent can push into Apimo

Five writes matter to an agent that answers and qualifies enquiries. The two that need the most care are the lead, which updates by its reference, and the contact, which gets duplicated if the agent does not search first. The agency side of this flow is on our page about buyer qualification by an AI agent.

One detail in the spec prevents a calendar mistake. Whenever the agent sends an action's type, updates included, the calendar display flag (calendar_is) goes back to its default, unless the agent sends it again in the same request. An edited viewing can drop out of the negotiator's calendar that way.

What the agent cannot do through the API

It cannot delete a contact, a lead or an action: the three documented deletes cover a listing's photos, media and documents. It sends no email or SMS, because there is no sending operation. Nothing alerts it to a change, because there is no webhook. It cannot read a negotiator's free slots: no route returns them, so the agent works them out from the actions it reads in each user's calendar.

Nor can it write proof of consent. Per-channel consent (phone, email, SMS, messaging) can be read on the contact record, but the contact write schema does not include it. The consent section below covers where to keep that proof.

How follow-ups get triggered

Apimo never alerts the agent, so every follow-up starts with a read. On each pass the agent compares the dates the API exposes with today's date, and drafts a message when a threshold agreed with the agency is crossed. The full use case is on our page about follow-ups with buyers and sellers.

The sending itself goes through a third-party tool, for example an n8n workflow wired to the agency's mailbox or messaging, or through Automator, Apimo's own automation module. Apimo's announcement of 12 December 2024 says Automator 2 sends email, SMS and push reminders to the negotiator or the client, synced with Google or Microsoft Calendar, with fully custom scenarios announced for later. Once a message has gone out, the agent logs it as an action on the contact record, so the negotiator sees it in the history.

Who signs off on what

Before go-live, the agency puts three things in writing: which rights the account gets (read first, then write, one object at a time), which messages go out without review, and which negotiator receives what the agent does not handle. We recommend that a negotiator approves every follow-up for the first few weeks, then only those that mention a price, a commitment date or a mandate.

Anything touching negotiation, price or signature stays with an authorised person. Our guide to the Hoguet law and AI agents sets out, article by article, what an agent can say on its own.

Hektor. On 29 October 2026, the product page shows "from 79 euros excl. VAT per month" for Hektor alone and more than 8,500 agencies equipped, with no API documentation. Key-based connections exist: MyNotary's page describes four steps, one of which is entering a MyNotary key into Hektor. The partner supplies the key and the software accepts it. Nothing shows that an agency can get the same access for its own agent.

Immofacile. The Orisha Real Estate page lists listing syndication, an electronic mandate register, automated buyer follow-ups and an in-house AI assistant called Scout. It describes the software as "compatible avec les autres logiciels immobiliers et solutions d’Orisha Real Estate" (compatible with other real estate software and Orisha Real Estate products) and says it connects to property management software. No published price, no API documentation. The partner page invites vendors to get in touch.

Modelo Office. On 29 October 2026, Modelo's website presents six integrated products and 19,000 client agencies. The pricing page holds only a form. It carries no API documentation.

What this means for an AI agent: on these three products it can work upstream (reply, qualify, prepare the record), but writing into the software depends on an agreement with the publisher. Before signing with a provider who promises "integration with your software", ask what access they have obtained, and in what form.

Twenty is an open-source CRM founded in Paris. Its documentation describes REST and GraphQL APIs generated from each workspace's data model: a "Listing" or "Viewing" object created by the agency gets its own endpoints. Authentication uses an API key sent as a Bearer token, and the webhooks page describes a call on every create, update or delete, signed with HMAC SHA256. The core is free to self-host. In the cloud, on 29 October 2026, the pricing page shows 9 dollars per user per month (Pro) and 39 dollars (Organization) billed yearly, with 50 and 100 API calls per minute.

HubSpot exposes its CRM objects through an API with OAuth 2.0. Its documentation shows how to create a meeting associated with a contact with POST /crm/v3/objects/meetings, and webhook requests carry a v3 signature computed with the app secret. HubSpot has no standard Listing object: listings have to be modelled separately. On 29 October 2026, the pricing page shows a free plan for up to two users and Starter from 7 dollars per seat per month.

The choice between the two comes down mostly to hosting. Twenty installs on a European server the agency controls; HubSpot is hosted by the vendor. What we do on Twenty is described on our page Twenty CRM installation and customisation.

Reading listings is what separates sector software from generic CRMs. In Apimo, the listing is the central object, with its mandate dates and its matches with buyer searches. In Twenty or HubSpot, the object has to be created and kept up to date, which means syncing with the software that syndicates the listings. That is the set-up described on our page on the intelligent CRM for real estate.

Two design rules apply whatever the software. The agent starts read-only, then gets write access object by object. And every write leaves a trace the negotiator can read: a note saying what the agent did, at what time, from which message.

The full path of a request, from the portal email to the viewing slip, is set out in our guide on how an estate agency can automate viewing requests with an AI agent.

Article 13 of law n° 2025-594 of 30 June 2025 rewrote article L. 223-1 of the French Consumer Code: the professional bears the burden of proving that the consumer's consent was collected. The table below turns articles R. 223-1 and R. 223-2 of the decree into CRM fields. What the law changes for finding sellers is covered in our guide to France's cold-calling ban for estate agencies.

Here the Apimo spec shows a detail none of the comparisons we read mention. The contact record exposes, for reading, consent per channel (phone, email, SMS, messaging), each with a consent or refusal and a date. These fields are not in the contact write schema, and none keeps the wording shown or the duration. An agent connected to Apimo can therefore check phone consent before offering a callback, but it must write the proof elsewhere: a dedicated field if the publisher opens one, or a separate log linked to the contact ID. In Twenty, a "Consent" object with its date and text fields is created like any other custom object.

The agent itself must follow the rule: it only offers a phone callback to a prospect whose consent is recorded and valid, and it records the withdrawal when the prospect asks for it in the conversation. Why the contact base should remain the agency's property is the subject of our guide on why estate agencies must own their data.

Lumyniq installs Twenty: bear that link in mind when reading this grid.

In our view, switching software just to connect an agent is rarely a good reason. A migration affects listing syndication, mandate history and the habits of the whole team. A written access request to the publisher costs far less, and the answer already tells you how far an agent can go. For agencies that also manage rentals, property management software and its automation are compared in our guide to automating property management with AI.

Record quality. An agent reads what the negotiators entered. Sold listings still marked active, duplicate contacts or buyer searches without criteria produce confident wrong answers. Cleaning the base therefore comes before connecting.

The old file. Contacts collected before 11 August 2026 without compliant consent do not become callable because an agent reads them again. Consent has to be collected again in the form article R. 223-1 sets: a clear request, a positive act, a duration of one year at most.

The processor. The agent, its host and the language model provider process your clients' personal data. Each must appear in the record of processing activities, with a contract.

The date. This check is valid for 29 October 2026, and the Apimo spec changed that same day. A publisher that publishes nothing today may open an API tomorrow, and a public API may change its terms. Check the documentation on the day you start the project.

On Twenty, we deliver installation in the cloud or on a European server, the business objects, workflows (follow-up after a viewing, end-of-mandate alert), data migration from another CRM and the connection to Claude. On existing software such as Apimo, we build a custom AI agent connected to its API, once the agency has activated the connection. For software with no documented API, we say so before quoting. The overview of AI projects in the sector is on our page AI for real estate.