destill.ai · All sectors · Deutsch · Trust Center

For banks, insurers and anyone who runs AI agents

Try it in four steps: change one URL

Your application keeps its OpenAI-compatible client. You point it at our gateway, every call goes through our prompt-injection screen before the model and an output check after it, and you see what happened. All four steps below are live as a pilot service without an SLA; where there is a limit, it says so.

Request a pilot keySee the code

1. Request a pilot key from us

Use the pilot form below, or write to hello@destill.ai, with your organisation and what you want to try. We reply with a key for your own pilot tenant. There is no self-service sign-up: each key is issued by a person, and each key only ever sees its own tenant.

2. Change the base URL

Set the base URL to https://api.destill.ai/v1 and send your key as a Bearer token. The gateway speaks the OpenAI chat-completions format (/v1/chat/completions, also streamed). Nothing else in your code changes.

curl

curl https://api.destill.ai/v1/chat/completions \
  -H "Authorization: Bearer $DESTILL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "qwen2.5:1.5b", "messages": [{"role": "user", "content": "Say hello in one sentence."}]}'

Python (openai package)

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.destill.ai/v1", api_key=os.environ["DESTILL_API_KEY"])
reply = client.chat.completions.create(
    model="qwen2.5:1.5b",
    messages=[{"role": "user", "content": "Say hello in one sentence."}],
)
print(reply.choices[0].message.content)

Node (openai package)

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://api.destill.ai/v1", apiKey: process.env.DESTILL_API_KEY });
const reply = await client.chat.completions.create({
  model: "qwen2.5:1.5b",
  messages: [{ role: "user", content: "Say hello in one sentence." }],
});
console.log(reply.choices[0].message.content);

The model: qwen2.5:1.5b is the model behind the pilot gateway today. The model is configurable per pilot; we agree it with you. Without a valid key the gateway answers 401. A prompt the screen stops is refused with the error prompt_blocked instead of reaching the model. The gateway is a pilot service without an SLA; if a call answers 404, it is not switched on publicly yet: write to us.

3. Watch your dashboard

Open api.destill.ai/v1/dashboard and paste your key. The dashboard shows only your own calls: what the screen let through, what it stopped, and errors, minute by minute, live as they happen. The key stays in your browser tab. Limit: the dashboard holds the calls since the gateway last started; the full record of your pilot is the signed report in step 4.

4. Download your signed report

Download a report of what the gateway saw and did for a time range, signed by us (did:web:destill.ai), so that you can check it without trusting us. It holds counts and hashes of your calls, never the prompts themselves.

# signed report (evidence record, JSON); without from/to: today so far, at most 31 days
curl -H "Authorization: Bearer $DESTILL_API_KEY" \
  "https://api.destill.ai/v1/report?from=2026-11-01&to=2026-11-30" -o report.json

# the same report as a readable page
curl -H "Authorization: Bearer $DESTILL_API_KEY" \
  "https://api.destill.ai/v1/report?format=html" -o report.html

# check it: the answer says "ok": true; one changed byte in the content makes it fail
(printf '{"record":'; cat report.json; printf '}') | curl -X POST -H "Content-Type: application/json" \
  --data-binary @- https://api.destill.ai/v1/verify

The check needs no key. Our public key is published under did:web:destill.ai, so you can also check the signature with your own tools. A key only ever gets its own tenant's report. Gateway calls are counted from 30 October 2026 on. The pilot gateway signs with our production key, not a test key; a separate sandbox (sandbox.api.destill.ai) exists for experiments, and its records always verify as "SANDBOX – not evidence".

The pilot offer

Scope and duration

One application of yours, one pilot key, one tenant. Four weeks from the first call; we can extend by agreement. You change one base URL, nothing else.

Success criteria, agreed in writing at kick-off

  • Every call your application sends in the pilot window appears in your dashboard and in the call count of your signed report.
  • Your team checks the signed report on its own, with /v1/verify or its own tools, and sees one changed byte fail.
  • The extra time per call is measured on your own calls; you decide whether it is acceptable.

These are what we commit to measure with you, not results from an earlier pilot.

Price logic

The pilot has a fixed price, agreed after a first call about your scope. After a successful pilot: an annual licence, scaled by volume. Amounts only in the written offer.

Contracting party: destill.ai Ltd. (Cyprus, in formation). Until it is registered, Lebensfluss e.V. (Austria, see footer) signs the pilot.

Hosted or in your own environment

Hosted (default): your calls pass through our server in the EU (Hetzner, Germany). For a bank or insurer that makes us an ICT third-party service provider under DORA (Regulation (EU) 2022/2554, Art. 28 ff.); we answer your third-party questionnaire as part of the pilot. Source: eur-lex.europa.eu

In your own environment: the receipt service (signed records, check, report) runs as Docker containers with its own Postgres database in your data centre. The setup is written down step by step and has been run end to end from nothing in a throw-away environment; you create your own signing keys, which never reach us. The gateway with the prompt screen is in the same Docker project, but its setup in your environment is not yet a written, tested path: we do that with you during the pilot.