Create your space
Signing up takes your email address and a name for your workspace. You receive a confirmation link by email, valid for 30 minutes and single-use. The other form fields (company, sector, size) are optional and gate nothing.
Once in, you choose where your Kastel runs: hosted by Kastel, or hosted by you. This section documents the first path. Kastel hosting is an isolated instance, for your company and nobody else, operated in Europe.
You then pick your plan and register your card: the subscription starts immediately, with no commitment, and can be cancelled anytime from your console. There is no hosted free trial: Kastel’s trial path is the free self-host. Plans, prices and included seats are on the Pricing page.
Healthcare, finance, public sector, industry with sensitive IP: the paid plan that concerns you is Enterprise, on quote, with no online checkout. Write to us, we answer the same day.
Write to usAfter payment, your space starts building on its own and takes a few minutes. You can leave the page open: it takes you to the next step as soon as your space is ready. Have an account with an AI provider at hand: your Kastel will ask you for a key right after.
Connect your AI keys
Your Kastel works with your company’s own AI keys: you hold the account with the AI provider, and the provider bills you, never Kastel. This step is mandatory: without a key, your Kastel can neither read nor build anything.
Anthropic, OpenAI, Mistral, DeepSeek or Kimi, each with its jurisdiction shown. Already have an enterprise cloud account? You can plug in your own Amazon Bedrock or Google Vertex account, in a European region.
If you set one, your Kastel stops calling the AI for reading and building once the month’s cap is reached. Two uses stay metered separately, outside the cap: search, and lesson compilation. Change the cap whenever you want.
It is sent directly to your space, where it is stored encrypted. It is never kept anywhere else.
A second key, dedicated to finding the right information in your Kastel. It comes from Mistral (France), whatever provider you chose at step 1: if you already chose Mistral, the same key works here; otherwise, allow five minutes to create a Mistral account.
Connecting your AI keys gives your Kastel what it needs to read and build: the AI provider bills you for that work. Giving an AI access is something else: it authorises an external AI (Claude, Cursor, ChatGPT...) to consult your Kastel. That second gesture is covered by the Connect your AI guide.
Read the Connect your AI guideChoose what your Kastel reads
The principle that governs everything else: the AI only ever passes over what a human designated. Your Kastel is not a mirror of your tools, it is a chosen corpus. Each department head decides their own scope, and nothing launches without a quote. The first two steps below chain on their own right after your keys are connected.
The documents that describe your company: org chart, company deck, offer, master procedures... A checklist guides you; nothing blocks, it indicates, it never demands.
Your Kastel asks you about your organisation, building on what you dropped. Started on one computer, the interview resumes anywhere. At the end, you confirm your organisation map: you are the one who confirms, and that confirmation is recorded under your name.
Your departments are the ones you just confirmed on your organisation map. Open one, then its Sources tab: pick a connected source (Google Drive, Slack or Notion), your Kastel lists the folders and channels it holds without opening a single document and without spending anything, and you tick what matters.
The quote is always a range in euros, with what remains of your monthly cap if you set one. If the high end exceeds the cap, the page says so before offering the button: trim your selection, or raise your cap. At the end, a report details what was read, what was skipped, and what it cost. What has already been read and has not changed is never re-read nor paid twice - the quote itself stays cautious and does not deduct that saving.
Turn on what runs on its own
By default, nothing turns itself on: no function that spends your AI keys starts without an explicit gesture from you. That is a sovereignty choice, not a saving: an AI that re-reads your context on its own initiative must have been told to, by you.
The counterpart is stated just as plainly: while these functions are off, your Kastel stays heavily degraded, it learns nothing and watches nothing. At the end of the build, it recommends turning them all on. The six:
Every week, re-reads the pages that could contradict each other and flags real conflicts. Uses your AI keys.
Checks right away that new content contradicts nothing existing. Uses your AI keys.
Remembers who said what and with what certainty. The toggle itself spends nothing; reading the statements is launched separately, quote shown.
A mechanical, free check, with no AI call at all.
Distils what your agents learned into reusable lessons. The function whose switch-off degrades your Kastel the most. Uses your AI keys, outside your monthly cap.
Re-measures the Context Score every week so the figure shown stays fresh. Quoted before every run, and the measurement only starts from ten labelled questions: below that, nothing runs and nothing is spent.
Everything is driven from Settings then Automations, the “What runs on its own in your space” page: turn on, turn off, and read the month’s estimated spend for the functions that are metered. Only the owner or an administrator can toggle, and every toggle is recorded in your audit trail.
Plug in your AI
Last gesture of the first day: give your AI access to your Kastel. Each person connects their own AIs, the authorisation carries their name, and can be cut at any time from “Your connected AIs” (an already-open connection can stay active for up to a quarter of an hour). The Connect your AI guide gives the ready-to-paste connection block for Claude, Cursor, ChatGPT and any MCP client.
Your Kastel is built, your keys are in place, your AI is plugged in. The next step is inviting your team: people, roles, and a governance that records who decides what. You decide what each AI is allowed to see.
Read the Connect your AI guide