Pricing
No per-seat licence, no minimum commitment, no charge for the seats that only read dashboards. A modest platform fee buys throughput and support; everything else is usage, billed to the token in arrears.
For prototypes, side projects and the evaluation you have to run before anyone will sign anything.
5M tokens included every month
Start freeFor the first agent that real people depend on, with throughput to survive a launch.
Includes €40 of usage credit
Start 14-day trialFor teams running several agents in production, with compliance people asking questions.
€350 credit and 15% off every token
Talk to usPrivate deployment in your own VPC, a signed DPA with the sub-processors you approve, custom retention, and a named engineer who answers in an hour. Annual contracts start at €30,000.
Cost calculator
Move the sliders to match your workload. The estimate uses real published rates, picks the cheapest plan for you, and shows the arithmetic rather than hiding it.
Input and output rate per million tokens.
System prompt, conversation history, tool definitions and any retrieved passages.
Repeated prefixes cost 90% less. Agent workloads typically land between 60% and 80%.
Estimated monthly cost
€0
Cheapest on the Starter plan.
That is €0.0000 per request.
Start on the free tierEstimates exclude VAT and embedding costs at index time. Your first month is capped automatically — you cannot overspend by accident.
Included everywhere
Nothing you send is used to train a model, on any plan, including the free one. Zero-retention mode is a single flag.
Cost per run, per key and per collection, live. Export the raw ledger as CSV whenever finance asks where the money went.
Set a monthly ceiling and Lumen refuses new runs past it. No overage invoice, no apologetic email from us.
Fine print
A token is roughly four characters of English, or about ¾ of a word. We count both directions: the text you send and the text the model generates.
Tool definitions, system prompts and retrieved passages are input tokens too. That surprises people, which is why the calculator above asks for total input rather than just your prompt.
New runs return 429 budget_exceeded and nothing is billed beyond the cap. Runs already in flight finish normally so you never get a half-written answer.
You can raise or remove the cap instantly from settings, and alerts fire at 50%, 80% and 100% by default.
Embedding at index time is billed per token on lumen-embed — €0.06 per million, so indexing a 500-page handbook costs under a cent.
Queries against an existing index are free. The passages that come back are billed as input tokens on the run that used them.
Yes, both directions, and the platform fee is prorated to the day. Usage credit is prorated the same way, so upgrading mid-month never costs you the credit you already had.
Two months free on annual platform fees. Usage stays monthly in arrears at the same per-token rate — we will not ask you to pre-buy credits that expire.
One call to POST /v1/runs, however many tool calls happen inside it. An agent that makes nine tool calls and streams for forty seconds is still one request.
Registered non-profits and companies under two years old with less than €2M raised get €5,000 of credit and Scale-tier limits for twelve months. One short form, no pitch deck.
One button in settings, effective at the end of the billing period, no retention call. Export your collections and run history first — both stay available for 30 days after you leave.
Five million tokens a month is genuinely enough to launch something small. When it is not, the next step is €49.