Conversation channels
Your AI agent · all of Switzerland
Your agent,
your own.
You give it a goal. It reads, searches, prepares and monitors. You keep the decisions that matter.
Everything your agent can connect to
Your tools. Your channels. Your choice.
This map starts from the public catalogs and the MCP registry, the open protocol that connects an agent to tools. Open a category to see the names checked on 8 October 2026.
An available connection is not an authorization. Next, the accounts, folders and permitted actions are chosen, along with the ones that need your approval.
Email10 names
Read, sort, search, prepare drafts and send after approval.
- Gmail
- Microsoft Outlook
- Fastmail
- Proton Mail
- Yahoo Mail
- IMAP
- SMTP
- Microsoft Exchange
- SendGrid
- Mailchimp
Sources: Google Workspace MCP servers · Microsoft Graph · Zapier MCP — catalog of 9.000 applications
Calendar and appointments8 names
Read availability, suggest a time slot, create or edit an event.
- Google Calendar
- Outlook Calendar
- Apple Calendar
- CalDAV
- Calendly
- Cal.com
- Microsoft Bookings
- Zoom
Sources: Google Workspace MCP servers · Microsoft Graph · Zapier MCP — catalog of 9.000 applications
Files and documents12 names
Search, read, create, file, convert and share according to the permissions given.
- Google Drive
- Google Docs
- Google Slides
- Dropbox
- Microsoft OneDrive
- Microsoft SharePoint
- Microsoft Word
- Box
- local files
- S3
- Nextcloud
Sources: Google Workspace MCP servers · Microsoft Graph · Reference MCP servers
Spreadsheets and lightweight databases8 names
Read rows, calculate, reconcile, write results and produce exports.
- Microsoft Excel
- Google Sheets
- Airtable
- Smartsheet
- CSV
- TSV
- Baserow
- Coda
Sources: Google Workspace MCP servers · Microsoft Graph · Zapier MCP — catalog of 9.000 applications
Notes and knowledge8 names
Query the notes, find a procedure and keep the source passages.
- Notion
- Confluence
- Obsidian
- Microsoft OneNote
- Evernote
- Slite
- Outline
- Readwise
Sources: Zapier MCP — catalog of 9.000 applications · Pipedream — application catalog · n8n — integration catalog
Team messaging8 names
Read an authorized thread, prepare a reply, post or notify.
- Slack
- Microsoft Teams
- Discord
- Google Chat
- Mattermost
- Telegram
- Rocket.Chat
Sources: Slack messaging API · Microsoft Teams bots · Discord bots
CRM and sales8 names
Look up a contact, enrich a record, track a deal and prepare a follow-up.
- Salesforce
- HubSpot
- Pipedrive
- Zoho CRM
- Microsoft Dynamics 365
- Close
- Copper
- Freshsales
Sources: Zapier MCP — catalog of 9.000 applications · Pipedream — application catalog · n8n — integration catalog
Accounting, invoices and payments10 names
Extract an invoice, match a payment and prepare a reminder or an export.
- QuickBooks
- Xero
- Stripe
- PayPal
- Square
- FreshBooks
- Sage
- Odoo
- Bexio
- Pennylane
Sources: Stripe API · Zapier MCP — catalog of 9.000 applications · Pipedream — application catalog
E-commerce9 names
Read products, orders, stock and customers; prepare an update for approval.
- Shopify
- WooCommerce
- Adobe Commerce
- BigCommerce
- Amazon Seller
- eBay
- Etsy
- PrestaShop
- Wix eCommerce
Sources: Shopify API · Zapier MCP — catalog of 9.000 applications · Pipedream — application catalog
Design, images, audio and video9 names
Read a design, generate a draft, export a visual or process a media file.
- Figma
- Canva
- Adobe Creative Cloud
- Blender
- Miro
- ElevenLabs
- YouTube
- Vimeo
- Cloudinary
Sources: Figma MCP server · Zapier MCP — catalog of 9.000 applications · Pipedream — application catalog
Code, repositories and infrastructure14 names
Read a repository, open a ticket, run tests and prepare a fix.
- GitHub
- GitLab
- Bitbucket
- Jira
- Linear
- Sentry
- Datadog
- Docker
- Kubernetes
- Cloudflare
- Amazon Web Services
- Microsoft Azure
- Google Cloud
- Vercel
Sources: Official GitHub MCP server · Reference MCP servers · Zapier MCP — catalog of 9.000 applications
Browser and web automation8 names
Open a page, fill in a form, capture and check a result.
- Google Chrome
- Microsoft Edge
- Playwright
- Puppeteer
- Selenium
- Browserbase
- Browserless
- Apify
Sources: Official MCP registry · Reference MCP servers · Zapier MCP — catalog of 9.000 applications
Search and monitoring9 names
Search the web, read pages, compare sources and watch for a change.
- Brave Search
- Bing Web Search
- Google Programmable Search
- Tavily
- Exa
- Perplexity
- Firecrawl
- RSS
- Wikipedia
Sources: Official MCP registry · Reference MCP servers · Zapier MCP — catalog of 9.000 applications
Maps, places and travel8 names
Find a place, calculate a route and prepare a trip.
- Google Maps
- Mapbox
- OpenStreetMap
- HERE
- TomTom
- Amadeus
- FlightAware
- OpenWeather
Sources: Official MCP registry · Zapier MCP — catalog of 9.000 applications · Pipedream — application catalog
Home and devices8 names
Read sensors, watch a threshold and control only what is authorized.
- Home Assistant
- MQTT
- Matter
- Philips Hue
- Samsung SmartThings
- Sonos
- Tado
- Tesla
Sources: Home Assistant API · Official MCP registry · Zapier MCP — catalog of 9.000 applications
Health and sport7 names
Import authorized measurements and prepare trends, without diagnosis.
- Apple Health
- Google Health Connect
- Fitbit
- Garmin Connect
- Strava
- WHOOP
- Oura
Sources: Zapier MCP — catalog of 9.000 applications · Pipedream — application catalog · n8n — integration catalog
Finance and markets8 names
Read authorized accounts or prices, reconcile and alert without deciding on a purchase.
- Plaid
- Wise
- Revolut Business
- Alpaca
- Alpha Vantage
- CoinGecko
- Yahoo Finance
- Interactive Brokers
Sources: Official MCP registry · Zapier MCP — catalog of 9.000 applications · Pipedream — application catalog
Databases and analytics10 names
Query, transform and summarize data under limited permissions.
- PostgreSQL
- MySQL
- SQLite
- Microsoft SQL Server
- MongoDB
- Redis
- Elasticsearch
- Snowflake
- BigQuery
- Databricks
Sources: Reference MCP servers · Official MCP registry · Zapier MCP — catalog of 9.000 applications
Projects and customer support10 names
Create or update a task, a ticket, an incident or a support reply.
- Asana
- Trello
- monday.com
- ClickUp
- Basecamp
- Zendesk
- Intercom
- Freshdesk
- ServiceNow
- PagerDuty
Sources: Zapier MCP — catalog of 9.000 applications · Pipedream — application catalog · n8n — integration catalog
Automation and APIs8 names
Connect anything that has an API, a webhook, a command or an MCP server.
- MCP
- REST API
- GraphQL
- webhooks
- Zapier
- Make
- n8n
- IFTTT
Sources: Official MCP registry · Reference MCP servers · Zapier MCP — catalog of 9.000 applications
ERP, HR and operations8 names
Read an authorized file, prepare an update and follow an internal process.
- SAP
- Oracle NetSuite
- Odoo
- Microsoft Dynamics 365
- Workday
- BambooHR
- Personio
- DocuSign
Sources: Zapier MCP — catalog of 9.000 applications · Pipedream — application catalog · n8n — integration catalog
Model providers
Its engine can change.
The public sources are listed in each category. The MCP registry keeps changing: a connection must always be checked before use.
What an agent does today
It does more than answer.
With the right access, an agent can chain several steps. It opens a tool, looks up information, prepares a result and stops when your approval is needed.
An AI on its own
It answers a request.
It writes, explains or summarizes what you give it. Without a tool, it does not open your mailbox and does not know your files.
An agent
It prepares a task from start to finish.
It uses only the authorized tools and folders. It can repeat a check, keep a record and ask you before acting.
30 concrete examples
Personal life
Your inbox is overflowing
It spots the important messages, groups them, summarizes what needs to be done and prepares the replies without sending them.
A form is waiting for you
It reads the documents you provide, pre-fills the fields, flags what is missing and leaves you to check and then submit.
You are waiting for a price drop
It checks the pages you chose, compares prices and delivery, then alerts you without buying.
You need to set up an appointment
It checks the authorized calendar, finds time slots and prepares the invitation. You approve the sending.
A subject is hard to understand
It looks for dated sources, compares the explanations, summarizes the solid points and keeps the links.
Your files are a mess
It suggests names and folders, shows the planned move, then files them after your approval.
Home and family
Calendars overlap
It gathers the authorized calendars, spots conflicts and suggests time slots without moving an appointment on its own.
Meals get decided too late
It starts from your preferences and the pantry you declared, suggests a few menus and prepares an editable shopping list.
Invoices and warranties get lost
It picks out the useful dates, files the documents entrusted to it and prepares the reminders for you to check.
Maintenance falls through the cracks
It reads the available manuals, prepares a list and reminds you of the next due date.
A device consumes too much
It reads the authorized measurements, compares them with the chosen threshold and flags the gap without controlling the device.
You are planning an event
It gathers availability and constraints, proposes a schedule and writes the messages for you to approve.
Self-employed
A customer asks for a quote
It reads the request, spots the missing details and prepares the quote from your rates. You approve.
Receipts pile up
It extracts the date, supplier and amount, suggests a category and sets aside what needs checking.
Invoices are unpaid
It compares due dates and recorded payments, draws up the list and prepares a suitable reminder.
The decisions from a call get lost
From notes or an authorized recording, it extracts decisions, owners and dates.
Your website is out of date
It finds the information concerned, proposes the new text and prepares a preview before publishing.
A small piece of software has a bug
It reproduces the problem in a copy, proposes a fix, runs the tests and shows the changes.
Small business
The same questions keep coming back
It searches your approved answers, prepares the message and passes uncertain cases to a person.
An invoice looks odd
It matches the order, delivery and invoice, notes the discrepancies and prepares a query without paying.
Stock-outs catch you by surprise
It reads the available stock, checks the thresholds and prepares an order proposal for you to confirm.
Appointments keep interrupting the team
It picks up the request, checks the authorized slots and suggests times according to your rules.
Communication keeps getting put off
It turns the approved information into a newsletter or posts, then leaves it to you to publish.
A piece of equipment sends too many alerts
It groups the signals, explains the anomalies observed and alerts the right person.
Larger company
The right procedure cannot be found
It searches the documents the person has access to, answers with the source passages and flags contradictory versions.
Offers need to be compared
It extracts the same criteria, prepares a table and notes the exclusions. The people in charge decide.
A new colleague is arriving
It prepares the welcome list, gathers the authorized documents and asks for each access to be opened.
The monthly report takes days
It gathers the exports, reconciles the totals, flags inconsistencies and prepares a verifiable report.
IT alerts are overflowing
It groups similar events, consults the approved procedures and escalates without touching production.
Projects lack a shared view
It extracts decisions and blockers from the authorized meeting notes, then prepares the follow-ups for approval.
Why have your own
One agent. Several doors.
The model is its engine for writing and reasoning. Your agent adds your tools, your rules and your habits. You can change engine without starting again from a blank page.
On your side
A local model
Documents can stay on your machine. It must be powerful enough, switched on at the right time and actually tested on your task.
Already paid for
A subscription
Your subscription can sometimes be used through an official connection. It keeps its quotas and its rules. It does not become a universal API.
Pay per use
A per-token provider
You pay for the text received and produced. A small model can handle the routine; a stronger model steps in when the task calls for it.
The model, like a power socket
Your rules stay visible.
Your tone, your authorized folders, your document templates and the decisions to ask you about belong to the agent’s configuration. A change of provider needs testing, but it should not erase all that work.
You also choose what is remembered, what must be corrected and what must be deleted.
One or several agents
A single one can connect everything. Several can share the work.
For a first task, one agent is often enough. As needs grow, each agent can keep a clear role, its own access and a simple way to hand over to the others.
For the self-employed
One agent prepares the replies to customers. Another gathers quotes, invoices and receipts.
For a family
One agent follows the calendars. Another files the documents and prepares the useful reminders.
For a company
Support, stock and reports each have their own agent. They share only what is necessary.
Start with a single task.
We compare your current way of working with the trial: result, checking time, cost, errors and ease of stopping. A simple rule can sometimes be better than an agent. That is also a good answer.
What it costs
Three ways to pay for intelligence.
A token is a small piece of text counted by the provider. An agent task can make several calls. The price of the model does not necessarily include the installation, voice, web search, hosting or checking time.
Local
No per-token bill for the open model cited. What remains is the hardware, electricity, setup and upkeep.
Subscription
A monthly amount, with limits. API access is generally separate and connections must be authorized.
Pay as you go
Input and output are billed separately. The real cost depends on the number of steps, the documents and the retries.
| Solution | Specific example | Price recorded | Date and source |
|---|---|---|---|
| Local model | Qwen3-4B, Apache 2.0 license | CHF 0 model usage feeHardware, electricity, installation and monitoring not included. | Checked on 18.10.2026 · official Qwen page |
| US subscription | ChatGPT Plus | USD 20 per monthAPI billed separately; usage limits. | Recorded on 18.10.2026 · official OpenAI page |
| US subscription | Claude Pro, billed monthly | USD 20 per monthSeparate API; usage limits. | Recorded on 18.10.2026 · official Anthropic page |
| US API | OpenAI GPT-6 Luna, short context | USD 0.10 input / USD 0.50 outputPer million tokens, standard rate. | Recorded on 18.10.2026 · OpenAI API |
| US API | Anthropic Claude Sonnet 5.5 | USD 2.00 input / USD 10.00 outputPer million tokens, standard rate. | Recorded on 18.10.2026 · official Anthropic page |
| Chinese API · DeepSeek | V4.1-Flash | USD 0.15 / 0.60 off-peakUSD 0.30 / 1.20 at peak hours; input / output, per million tokens. | Recorded on 18.10.2026 · official DeepSeek page |
| Chinese API · Qwen | Qwen3.5-Flash, Global offer, input ≤ 128.000 tokens | USD 0.029 input / USD 0.287 outputPer million tokens. | Recorded on 18.10.2026 · official Alibaba Cloud page |
| Chinese API · GLM | Z.ai GLM-5.3-Flash | USD 0.15 input / USD 0.50 outputPer million tokens. | Recorded on 18.10.2026 · official Z.ai page |
| Chinese API · MiniMax | MiniMax M3, standard, input ≤ 512.000 tokens | USD 0.30 input / USD 1.20 outputPer million tokens; displayed discount included. | Recorded on 18.10.2026 · official MiniMax page |
| Chinese API · Kimi | Moonshot Kimi K3 | USD 3.00 input / USD 15.00 outputPer million tokens, uncached input. | Recorded on 18.10.2026 · official Kimi page |
These prices are provider reference points, not Saphan rates and not a quality comparison. Taxes, regions, length thresholds, cache discounts and ancillary services can change the result. The source prevails.
Honest limits
Useful, yes. Infallible, no.
A well-written result can be wrong. A chain of actions can go astray. The safeguards must therefore be in the access rights and in the software, not just in a sentence given to the AI.
It can invent a fact.
We keep the original documents, cite the sources and check the important figures.
A website can block it.
A page layout changes, a sign-in expires or a check calls for a person. The agent stops instead of working around it.
A long task can drift.
We break the work down, set a maximum number of steps and ask for a human takeover if the result remains uncertain.
A small local model may not be enough.
We test it on the real machine and the real task. We then do not send the data elsewhere without your agreement.
A document can contain a false instruction.
The content it reads remains information. It never becomes authorization to send, pay or open another file.
A memory can become outdated.
You must be able to review, correct, delete and export what the agent remembers.
What it does alone. What it asks you. What it does not decide.
The kit or us
Two ways to have your own.
You deploy it with our free kit, or we set it up with you, in all of Switzerland.
With the kit, free
One file. Your starting point.
- Download the kit One file to keep in a folder reserved for your agent.
- Open this folder in Claude Code In the Code tab of the desktop app, ask it to read the kit and guide you.
- Give it its role With examples, you choose its job, its plan, its name, its character and the channel where it answers you.
- Prepare its installation The kit has the folder prepared for Hermes Agent. You approve the installation steps and the access, then you try a first message.
The kit guides you through eight steps. You keep your agent’s plan and identity in your folder. You choose the model and the connections: an installation on your side may use remote services depending on those choices.
Download votre-agent.md ↗
Markdown · 11.695 bytes · version 1.1, October 2026. · in French
The kit is free. Claude Code, the model you choose and the services you connect follow their own terms. What is possible depends on your configuration.
Official documentation: Claude Code on desktop · Hermes Agent and its messaging apps. Pages read on 18 October 2026.
With us
The setup: 2 hours, CHF 360.–
From your idea to an agent that answers, on your accounts and your machine. We define its first task, prepare the installation and run the first trial together.
In all of Switzerland.
To go further: Agent tuned for one business function · Training · An agent with a sealed head.
Your agent, in a few questions
What would you like to start with?
One sentence is enough to start. Long answers are welcome. You will read the whole message again before sending it through the secure form or opening WhatsApp.