DUKTUS PRO API

Therapeutic AI for practice management software. GDPR-compliant, EU-hosted in Frankfurt.

Chat — Therapeutic AI

Send plain text with patient names. The API auto-anonymizes, runs the AI, returns the answer with names restored. The AI model never sees PII.

8–20 ct per call

POST/chat

Voice — Audio Transcription

Transcribe session recordings to text. EU backend keeps audio in Frankfurt. Optional automatic PII anonymization of the transcript.

12–18 ct per call

POST/voice/transcribe

What's available

Service	Description	Price
Chat	Therapeutic AI with 5 modes — standard, diagnostics, planning, documentation, crisis	8–20 ct
Voice	Audio-to-text transcription with optional PII anonymization. EU backend available.	12–18 ct
Assessments	PHQ-9, GAD-7, PCL-5, ISI, AUDIT-C scoring with clinical references	Free
Anonymizer	GDPR PII detection and reversible anonymization	3 ct
Evidence	ICD-10 disorder database, clinical guidelines, interventions	0–10 ct
Generator	Clinical documents from 11 validated templates, PDF export	20 ct
Text Improve	Convert shorthand notes into professional clinical language. Uses `dkts_base_mini` — smallest, fastest model	6 ct

Base URL: https://api.duktus-pro.de/api/v1

DUKTUS PRO is built for healthcare. Every design decision prioritizes patient data protection.

Measure	Detail
EU Infrastructure	All services run in AWS eu-central-1 (Frankfurt, Germany)
Voice — EU option	`backend=private` routes audio through our EU transcription service in Frankfurt. Audio never leaves the EU.
Reversible anonymization	Encrypted mappings are returned with every response — you can restore names at any time using `/deanonymize`
Art. 9 compliance	Designed for health-category data processing under GDPR Article 9
No AI training on your data	Patient conversations are not used to train the underlying models
Audit trail	All API calls are logged per-partner with timestamps for compliance audits

Authentication

All endpoints except GET /health require an API key in the request header.

HTTP Header

X-API-Key: dk_live_YOUR_40_CHAR_KEY

Keys are issued by DUKTUS PRO. Format: dk_live_ followed by 40 characters. Contact api@duktus-pro.de to request access.

Never include your API key in client-side code or commit it to version control. Contact us immediately to rotate a compromised key.

First API Call — Health Check

No authentication required. Confirm the API is reachable and all services are operational.

bash

curl https://api.duktus-pro.de/api/v1/health

python

import requests
r = requests.get("https://api.duktus-pro.de/api/v1/health")
print(r.json())

javascript

const r = await fetch('https://api.duktus-pro.de/api/v1/health');
const data = await r.json();
console.log(data.status); // "healthy"

response

{
  "status": "healthy",
  "api_version": "2.1.5",
  "services": {
    "anonymizer": { "status": "healthy" },
    "evidence":   { "status": "healthy" },
    "generator":  { "status": "healthy" }
  }
}

Chat — Therapeutic AI

The primary endpoint. Send clinical context in plain text — PII anonymization is automatic. You send patient names, the AI never sees them.

Modes

Mode	Use case	Price
`standard`	General therapeutic questions	15 ct
`diagnostik`	ICD-10 differential diagnosis	20 ct
`therapieplanung`	Evidence-based treatment planning	20 ct
`dokumentation`	Session notes and clinical reports	8 ct
`krisenintervention`	Crisis assessment and safety planning	20 ct

bash

curl -X POST https://api.duktus-pro.de/api/v1/chat \
  -H "X-API-Key: dk_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "message": "Patient Maria Mueller, 34F, reports persistent insomnia for 3 weeks. PHQ-9: 12.",
    "mode": "diagnostik"
  }'

python

import requests

API_KEY = "dk_live_YOUR_KEY"
BASE    = "https://api.duktus-pro.de/api/v1"
headers = {"X-API-Key": API_KEY}

r = requests.post(f"{BASE}/chat", headers=headers, json={
    "message": "Patient Maria Mueller, 34F, persistent insomnia 3 weeks. PHQ-9: 12.",
    "mode": "diagnostik",
})

data = r.json()
print(data["response"])           # AI response -- names already restored
print(data["usage"]["cost_cents"])  # 20
print(data["session_id"])           # reuse for follow-up messages

javascript

const r = await fetch('https://api.duktus-pro.de/api/v1/chat', {
  method: 'POST',
  headers: { 'X-API-Key': 'dk_live_YOUR_KEY', 'Content-Type': 'application/json' },
  body: JSON.stringify({
    message: 'Patient Maria Mueller, 34F, persistent insomnia 3 weeks.',
    mode: 'diagnostik',
  }),
});
const { response, session_id, usage } = await r.json();
console.log(response);           // AI response (names restored)
console.log(usage.cost_cents);   // 20

Multi-turn conversations

Pass session_id from the first response to maintain context across follow-up messages:

python

r1 = requests.post(f"{BASE}/chat", headers=headers,
    json={"message": "Patient Max Bauer, PHQ-9 score: 18.", "mode": "diagnostik"})

session_id = r1.json()["session_id"]

r2 = requests.post(f"{BASE}/chat", headers=headers,
    json={"message": "What therapy approach would you recommend?",
          "session_id": session_id})

Voice — Audio Transcription

Transcribe session recordings to text. The private backend processes audio in Frankfurt (EU) — audio never leaves the EU. Enable anonymize=true to strip PII from the transcript.

Backends

Backend	Speed (1h audio)	Data location	GDPR	Price
`fast` (default)	~30–60 s	Outside EU	Not suitable for patient data	12 ct
`private`	~60–120 s	EU (Frankfurt)	GDPR-compliant	18 ct

Formats: WAV, MP3, WebM, M4A, FLAC, OGG (max 25 MB) · Languages: de, en, fr, es, it, pt, nl, pl, tr, ru, ja, ko, zh

bash

# Fast (outside EU -- for non-patient data)
curl -X POST https://api.duktus-pro.de/api/v1/voice/transcribe \
  -H "X-API-Key: dk_live_YOUR_KEY" \
  -F "audio=@recording.mp3" \
  -F "language=de" \
  -F "backend=fast"

# GDPR-compliant (EU Frankfurt) + auto-anonymize
curl -X POST https://api.duktus-pro.de/api/v1/voice/transcribe \
  -H "X-API-Key: dk_live_YOUR_KEY" \
  -F "audio=@recording.mp3" \
  -F "language=de" \
  -F "backend=private" \
  -F "anonymize=true"

python

with open("recording.mp3", "rb") as f:
    r = requests.post(
        f"{BASE}/voice/transcribe",
        headers={"X-API-Key": API_KEY},
        files={"audio": ("recording.mp3", f, "audio/mpeg")},
        data={
            "language":  "de",
            "backend":   "private",  # EU/GDPR
            "anonymize": "true",     # strip PII from transcript
        },
    )

d = r.json()["data"]
# d["transcript"]                     -> full text
# d["duration_seconds"]               -> audio length in seconds
# d["anonymization"]["entities_found"] -> PII removed (if anonymize=true)

Assessments

Clinical scoring tools — completely free. Validated against clinical standards.

Supported: phq9 (9 items) · gad7 (7 items) · pcl5 (20 items) · isi (7 items) · auditc (3 items) — all values 0–3 per item.

python

r = requests.post(f"{BASE}/assessments/score", headers=headers, json={
    "tool":      "phq9",
    "responses": [2, 3, 2, 1, 2, 3, 2, 1, 1],  # 9 items, 0-3 each
})

d = r.json()["data"]
# d["score"]     -> 17
# d["severity"]  -> "Moderate Depression"
# d["reference"] -> "Loewe et al. 2002"

Include assessment scores in your Chat messages for richer responses: "PHQ-9: 17/27 (Moderate Depression). Patient reports..."

Anonymizer

Detects and reversibly replaces PII: names, dates of birth, addresses, phone numbers, IBANs, insurance IDs, email addresses. The Chat endpoint does this automatically — call /anonymize directly only for custom pipelines.

python

# Anonymize
r = requests.post(f"{BASE}/anonymize", headers=headers,
    json={"text": "Patient Maria Mueller, DOB 15.03.1985, Berlin."})

anon = r.json()["data"]
# anon["anonymized_text"]    -> "Patient [PERSON_1], DOB [DATUM_1], [ORT_1]."
# anon["mappings_encrypted"] -> "eyJ..."  -- store this to restore names later

# Deanonymize
r2 = requests.post(f"{BASE}/deanonymize", headers=headers, json={
    "anonymized_text":    anon["anonymized_text"],
    "mappings_encrypted": anon["mappings_encrypted"],
})
# -> "Patient Maria Mueller, DOB 15.03.1985, Berlin."

Evidence Database

Browse ICD-10-GM disorders, clinical guidelines, recommended interventions, and assessment tools. GET endpoints are free; full-text search costs 10 ct.

python

# Paginated disorder list (free)
requests.get(f"{BASE}/evidence/disorders?page=1&per_page=20", headers=headers)

# Disorder detail by ICD-10 code (free)
requests.get(f"{BASE}/evidence/disorders/F32", headers=headers)

# Recommended interventions (free)
requests.get(f"{BASE}/evidence/interventions/F32", headers=headers)

# Full-text search (10 ct)
requests.get(f"{BASE}/evidence/search?q=depression", headers=headers)

Document Generator

Generate clinical documents from 11 validated templates — session notes, risk assessments, treatment plans, and more. PDF export available.

python

# List templates
requests.get(f"{BASE}/generate/templates", headers=headers)

# Generate document (20 ct)
r = requests.post(f"{BASE}/generate/document", headers=headers, json={
    "template_id": "verlaufsdokumentation_enhanced",
    "context": {
        "patient_info": "35M, F32.1",
        "session_number": 5,
        "content": "Patient reports improved sleep. Exposure exercises successful.",
    },
})
# r.json()["data"]["document"] -> ready-to-use clinical text

Text Improve

Convert rough notes and shorthand into professional clinical documentation. Uses dkts_base_mini — the smallest, fastest model (EU Frankfurt, GDPR Art. 44-49). Output is always in German clinical language regardless of input language.

Output styles

Style	Format
`session_notes`	Session note — neutral, descriptive, present tense
`report`	Assessment report — formal, structured, past tense
`letter`	Referral letter — very formal, full sentences
`diagnosis`	Diagnostic formulation with explicit ICD-10-GM codes
`general` (default)	General clinical language

python

r = requests.post(f"{BASE}/text/improve", headers=headers, json={
    "text":  "Pat. 34F, F32.1, 3 Monate, Schlafprob., Antriebslos.",
    "style": "session_notes",
})

d = r.json()["data"]
# d["improved"]  -> professional clinical text in German
# d["original"]  -> your input returned unchanged
print(d["improved"])

Pass the context field to hint at where the text will appear (e.g. "Treatment plan section 3: Goal definition") for better-calibrated output.

Extended Thinking

Add thinking: true to any Chat request to enable step-by-step reasoning. The AI works through the problem explicitly before responding — useful for complex diagnostic questions or treatment planning.

Parameter	Type	Default	Notes
`thinking`	boolean	`false`	Adds +15 ct surcharge
`thinking_budget_tokens`	integer	`10000`	1024 – 32000. Higher = deeper reasoning

python

response = requests.post(
    "https://api.duktus-pro.de/api/v1/chat",
    headers=headers,
    json={
        "message": "Patient shows F32.1 with somatic symptoms. PHQ-9: 18. GAD-7: 11. "
                   "Suggest a structured 12-session CBT plan including crisis protocol.",
        "mode": "therapieplanung",
        "thinking": True,
        "thinking_budget_tokens": 16000,
    }
)
data = response.json()

# Full reasoning trace (step-by-step)
print(data["thinking"])

# Final answer
print(data["message"])

bash

curl -X POST https://api.duktus-pro.de/api/v1/chat \
  -H "X-API-Key: $DUKTUS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "message": "Patient F32.1, PHQ-9 18, GAD-7 11. Suggest 12-session CBT plan.",
    "mode": "therapieplanung",
    "thinking": true,
    "thinking_budget_tokens": 16000
  }'

Cost: therapieplanung = 20 ct + 15 ct thinking surcharge = 35 ct total. Use standard mode (15 ct + 15 ct = 30 ct) for general thinking tasks.

Error Handling

All errors follow a consistent format with a machine-readable error.type and a human-readable error.message. Every response includes a request_id — include it when contacting support.

error response

{
  "type": "error",
  "error": {
    "type": "invalid_request_error",
    "message": "Field 'text' is required."
  },
  "request_id": "req_01abc..."
}

Status	Meaning	Action
`400`	Bad Request	Fix your request body
`401`	Unauthorized	Check your `X-API-Key`
`402`	Budget Exceeded	Contact DUKTUS to increase your budget
`404`	Not Found	Check endpoint path and resource ID
`422`	Validation Error	Check field types and allowed values
`429`	Rate Limited	Wait `Retry-After` seconds, then retry
`500`	Server Error	Retry once; contact api@duktus-pro.de with `request_id`

python — recommended error handling

import time, requests

def call_api(method, url, **kwargs):
    r = requests.request(method, url, **kwargs)
    if r.status_code == 429:
        time.sleep(int(r.headers.get("Retry-After", 60)))
        return call_api(method, url, **kwargs)  # retry once
    if r.status_code == 402:
        raise Exception("Monthly budget exceeded")
    r.raise_for_status()
    return r.json()

Rate Limits

Default: 100 requests/minute, 10,000/day per partner. Every response includes rate limit headers:

response headers

X-RateLimit-Limit-Minute:      100
X-RateLimit-Remaining-Minute:   87
X-RateLimit-Limit-Day:       10000
X-RateLimit-Remaining-Day:    9842
Retry-After:                     3  # only on 429

Pricing

Pay-per-use in EUR cents. Every response includes usage.cost_cents. HTTP 402 is returned when your monthly budget is reached.

Endpoint	Price per call
`POST /chat` (standard)	15 ct
`POST /chat` (diagnostik, therapieplanung, krisenintervention)	20 ct
`POST /chat` (dokumentation)	8 ct
`POST /assessments/score`, `/tools`	Free
`POST /anonymize`	3 ct
`POST /voice/transcribe` (fast — outside EU, not for patient data)	12 ct
`POST /voice/transcribe` (private — EU Frankfurt, GDPR-compliant)	18 ct
`GET /evidence/disorders/`, `/interventions/`, `/guidelines`	Free
`GET /evidence/search`	10 ct
`POST /generate/document`	20 ct
`POST /text/improve`	6 ct dkts_base_mini

Request API Access

DUKTUS PRO API

What's available

GDPR & Data Privacy

Authentication

First API Call — Health Check

Chat — Therapeutic AI

Modes

Multi-turn conversations

Voice — Audio Transcription

Backends

Assessments

Anonymizer

Evidence Database

Document Generator

Text Improve

Output styles

Extended Thinking

Error Handling

Rate Limits

Pricing