> ## Documentation Index
> Fetch the complete documentation index at: https://docs.llm.kiwi/llms.txt
> Use this file to discover all available pages before exploring further.
# Use cases
> Real-world LLM.kiwi use cases with prompts and patterns for chatbots, text summarization, data extraction, translation, and AI-assisted content generation.
LLM.kiwi can power server-side features that turn messages into generated text. Begin with a narrow task, test realistic inputs, and validate every result your application depends on.
Maintain relevant message history and provide a focused system instruction.
Convert long text into short bullets, action items, or structured notes.
Draft, rewrite, classify, or translate Croatian text with `hrllm`.
Ask for structured fields, then validate the returned format in code.
## Customer-support assistant
```python theme={null}
messages = [
{
"role": "system",
"content": (
"You assist customers using only the policy below. "
"If the answer is not in the policy, say you do not know.\n\n"
"Policy: Returns are accepted within 30 days for unused items."
),
},
{"role": "user", "content": "Can I return an unused item after 14 days?"},
]
response = client.chat.completions.create(model="auto", messages=messages)
print(response.choices[0].message.content)
```
Keep business rules in your own application as well. A model response should not be the only enforcement mechanism for refunds, payments, permissions, or other important actions.
## Summarize text
```python theme={null}
notes = """
The team agreed to launch on Friday. Ana will test checkout on Wednesday.
Marko will prepare the release notes. The remaining blocker is the email template.
"""
response = client.chat.completions.create(
model="auto",
messages=[
{
"role": "system",
"content": "Return a concise summary followed by a bullet list of owners and tasks.",
},
{"role": "user", "content": notes},
],
)
```
For long documents, stay within the model's context limit and split content into meaningful sections when needed.
## Generate Croatian text
```python theme={null}
response = client.chat.completions.create(
model="hrllm",
messages=[
{
"role": "system",
"content": "Piši prirodnim hrvatskim jezikom, jasno i bez marketinških fraza.",
},
{
"role": "user",
"content": "Preoblikuj ovu obavijest u profesionalan email: Paket kasni jedan dan.",
},
],
)
```
Review generated content before publishing it, especially names, prices, dates, legal statements, and factual claims.
## Extract structured data
```python theme={null}
import json
response = client.chat.completions.create(
model="auto",
messages=[
{
"role": "system",
"content": (
"Extract name and email. Return only JSON matching "
'{"name": string | null, "email": string | null}.'
),
},
{
"role": "user",
"content": "Contact Ana Horvat at ana@example.com.",
},
],
)
raw = response.choices[0].message.content
try:
result = json.loads(raw)
except json.JSONDecodeError:
result = {"name": None, "email": None}
```
A prompt asking for JSON does not guarantee valid JSON. Parse it, validate required fields and types, and define a fallback.
## Classify text
```python theme={null}
response = client.chat.completions.create(
model="auto",
messages=[
{
"role": "system",
"content": "Classify the message as billing, technical, or other. Output one label only.",
},
{"role": "user", "content": "My API request returns 401."},
],
)
label = response.choices[0].message.content.strip().lower()
allowed = {"billing", "technical", "other"}
if label not in allowed:
label = "other"
```
## Production checklist
* Call LLM.kiwi from your backend, not directly from a browser.
* Keep keys in environment variables and rotate leaked keys.
* Set request timeouts and handle API errors.
* Retry only temporary failures, with exponential backoff.
* Limit user input and conversation history.
* Validate structured output before using it.
* Log request IDs and failures without logging secrets.
* Add human review for high-impact actions or published content.
Prepare an integration for real users.
Copy complete language-specific examples.