No implementation fee. No per-seat tiers. No hidden AI token meter running against you at night. Fair-usage cloud AI included.
No credit card needed
No credit card needed
Every agent gets 20,000 credits a month. Credits track the model work actually done — not seats, not licences, not a meter that surprises you. Here is what real jobs cost.

@wildfire summarise yesterday’s #meridian thread and flag anything that needs me
Three items. Two are moving on their own: Volvo Q2 invoicing cleared, Bloom countersigned. One needs you — Nordsjö went quiet after the review. Reply drafted in HubSpot. Send it?

@wildfire the Nordsjö proposal still has Q4 numbers and the tone is off. Match it to the Meridian guide.
Done. Q1 market-share figures pulled from the vault, tone checked line by line against the guide. Two claims I could not source are flagged in the margin.
DNordsjö Rebrand 2026 · v4
@wildfire build us an approval desk — account leads submit discount requests, it pulls live deal data from HubSpot, partners approve in Slack
Live at deal-desk.meridian.dk. HubSpot connected, approvals post to #revenue, every decision signed into the ledger. 14 people invited.
| Deal | Amount | Status |
|---|---|---|
| Volvo Q2 | €48k | Pending |
| Nordsjö | €22k | Approved |
Credits follow real model usage. Caching brings them down. Nothing is hidden behind a tier.
Every credit reflects model work done — on Claude, GPT-5, Mistral, or the GPU in your own rack. No per-seat licence, no minimum commitment, no new platform tier appearing the month you grow.
Agents cache context and reuse what they already retrieved. The first Monday digest costs real credits. The fortieth costs a fraction of it.
Scheduled work runs on credits too, so hourly costs more than weekly. You set the schedule and you can see the meter — it is a line in the ledger, not a surprise on the invoice.
No credit card needed
Yes - 12 months upfront gets you 15% off the user portion. Talk to us if you need a multi-year option for budgeting.
A person who has their own named agent. People who only consume the team digest or read shared output don't count.
A unit of model work. Reading a thread, drafting a section, calling an integration — each carries a cost that reflects what the model actually did. A quick summary runs 100–300 credits. A built-and-shipped internal tool runs 2,000–5,000. Every agent gets 20,000 a month, included in the €99.
Yes, into the following month. They also pool across the workspace, so a quiet month in finance quietly covers a loud one in sales.
We tell you before you hit the wall, not after. Top up on demand, or let agents queue non-urgent work into the next cycle. Nothing stops without you deciding it should.
Yes, under fair usage - roughly 4 million tokens per user per month across Claude, GPT-5, Mistral. If you blow past that, we'll tell you before you see the next invoice.
That's the "Local Tier 3 GPU inference" add-on. We co-locate a dedicated Llama 3.3 70B node in the EU or run on your own iron. Pricing depends on traffic.
Annual term, 30-day exit notice, your data exports in standard formats on day one. The exit clause is in the DPA - read it before you sign.
No. SOC 2, ISO, DPA, evidence ledger, Article 14 exports - all in the base. Compliance isn't a tier.