# Explicio Grid | Sovereign AI batch API from Germany https://www.explicio.com/en/explicio-grid/ Last updated: 2026-11-02 > Explicio Grid accepts compute jobs in the OpenAI batch format and runs them on hardware Explicio operates in Germany. Billing follows delivered rows, the window is 24 hours on a best-effort basis, and the selected region is never left. PRODUCT · EXPLICIO GRID · DECENTRALISED AI COMPUTE NETWORK LOCATED IN GERMANY # Compute in batches – on hardware in Germany that we operate ourselves. Explicio Grid accepts large AI compute jobs in the OpenAI API batch format (JSONL) and runs them on the decentralised Explicio Boxes that we operate in Germany and the EU – without falling back outside the region you specify. No anonymous cloud routing, but a sovereign product with cryptographically walled-off boundaries: one job, one 24-hour window, one signed result file. Fair use rule: anything not delivered on time is not charged. Explicio Grid accepts compute jobs in the OpenAI batch format and runs them on hardware Explicio operates in Germany. Billing follows delivered rows, the window is 24 hours on a best-effort basis, and the selected region is never left. [Request API access](https://www.explicio.com/en/contact/?anliegen=grid) For Box customers: secure your 25 % discount ↓[Terms of use →](https://www.explicio.com/en/grid-nutzungsbedingungen/) - **Interface**: OpenAI Batch-compatible (files, batches, same error format) - **A job is**: An encrypted package (in version 1: embeddings) – nobody sees the plaintext - **Window**: 24 hours on idle time – a customer's own Box use always takes priority - **Region**: Fixed at submission (e.g. Germany) – no fallback elsewhere - **Billing**: Only delivered lines, charged to your billing account – failed lines cost €0 THREE STEPS · DROP-IN REPLACEMENT FOR OPENAI BATCH ## Upload. Submit. Collect. If you know the batch interface of the OpenAI API, you do not need to write any new code. The difference lies underneath: where, and how sovereignly, the computing is done. - UPLOAD 01 Job file in JSONL format One line per job. The file is strictly checked for format and model at the point of intake – anything that does not match the schema is rejected cleanly straight away and never enters the queue. - SUBMIT 02 Batch with a 24-hour window & fixed region You submit the batch with your choice of model and a fixed target region (e.g. Germany). As soon as free idle capacity is available, the job starts – until the 24-hour window expires, and never outside the region you chose. - COLLECT 03 Signed result file and error file Every valid line ends up in exactly one of the two files. Lines that were not processed within the 24-hour window appear transparently, with the reason, in the error file and cost nothing. EXPLICIO GRID · 60 SECONDS WHERE AND HOW COMPUTING HAPPENS · TECHNICALLY ENFORCED ## Six commitments written firmly into the code. LOCATION On hardware that we operate ourselves Jobs run exclusively on Boxes operated by Explicio. On customer Boxes only if they have actively consented to third-party workloads – and always first on our own fleet in Germany. PRIORITY The Box customer always comes first Grid jobs run strictly on idle time. As soon as the Box customer uses the Box themselves, the Grid job releases the hardware immediately – enforced technically in the scheduler, not a mere declaration of intent. PACKAGES Encrypted, signed, verified Every compute package is sealed to the key of the executing Box; the Box never sees any cloud credentials. The manifest is cryptographically signed, and a result is accepted only if it matches the manifest exactly. FAIR USE & TRANSPARENCY Pay only for delivery within the time window Because the Grid uses genuine idle time, we charge no premium for rigid real-time SLAs. Every line that is not finished within the 24-hour window is not charged. LIMITS Technical limits, stated openly Maximum file size, lines per file, tokens per line and concurrent jobs per account are clearly documented in advance. Anything that exceeds a limit is rejected at intake with a precise explanation. PROVEN The test before the promise Before the product came near a customer, it ran against thousands of disruptive Boxes and a batch of maximum size. Whatever broke was fixed in the kernel – the test report is publicly available in the documentation. THE OTHER SIDE · FOR OWNERS OF AN EXPLICIO BOX ## Anyone who books the Box Service saves 25 % of the monthly rate by computing along – voluntarily. 25 % DISCOUNT Idle time in exchange for a 25 % discount An Explicio Box that computes for the Grid on idle time at night and at weekends reduces your monthly service rate on the 48-month plan by 25 % straight away (e.g. from €1,220 to €915 / month for Tier 1). Participation is voluntary and can be switched off at any time. VOLUNTARY Two independent switches, not one Grid participation is never the default setting. Third-party jobs only reach a customer Box once the customer has actively consented in the admin portal and defined their own release windows. TRACEABLE Every contribution recorded in the Ledger What your Box has computed for the Grid is shown to the minute in the admin portal – per job, without any third-party plaintext content. The commercial credit follows the Ledger exactly, not a gut feeling. EXPLICIO GRID ## A decentralised compute network of boxes at our customers. Grid map — loads when it comes into view. Target picture: one cell per legal region. Jobs from the EU are computed in the EU cell. FREQUENTLY ASKED ## Before you ask us … How do I get access to Explicio Grid? + − So that the German Grid stays secure and enough compute capacity is available for every job, we personally activate company accounts for you after a short verification, including a data processing agreement (DPA). Which models run in the grid? + − In version 1 exclusively embedding models for the embeddings endpoint; which ones is stated in the model sheet of the terms of use. We do not offer text generation in the grid. What does a job cost? + − Billing is per delivered line according to the price list, monthly, in a separate invoice series. Every account has a credit limit; once the outstanding amount reaches the limit, the grid accepts no new jobs until payment is made. What happens if no capacity is free? + − The job waits within your region. Once the 24-hour window has passed, it expires: delivered lines are billed, the rest appear in the error file with a note and cost nothing. There is no separate notification about missing capacity — the monitor shows the situation. How long does my data stay in the grid? + − Job and result files are deleted after the retention period in the terms of use; a retrieval after that honestly answers “no longer available” instead of returning a copy. Nothing remains on executing boxes after a job. READ UP ## The terms are public. [Grid terms of use→Service, purposes, roles under the AI Act, retention, prices, credit limit — with the model sheet as annex A. German text.](https://www.explicio.com/en/grid-nutzungsbedingungen/) [Grid data processing agreement→For jobs that may contain personal data. German text.](https://www.explicio.com/en/grid-avv/) [Sub-processors→Who is involved besides us — and who is not.](https://www.explicio.com/en/unterauftragsverarbeiter/) ACCESS · AFTER ACTIVATION ## Give compute jobs to a place where it is clear where they run. Tell us which jobs you want to compute in batches. We will set up your account and reserve the exactly matching capacity window on our German hardware fleet for you. [Request access](https://www.explicio.com/en/contact/?anliegen=grid) [Explicio Box](https://www.explicio.com/en/explicio-box/) - **Format**: OpenAI batch (JSONL) - **Window**: 24 hours, best effort - **Region**: fixed, no fallback - **Billing**: delivered lines, credit limit per account