Integrate Mistral AI API: Large, Small, Codestral

We work with Mistral AI — a European LLM provider gaining traction in the enterprise segment. Its main advantage is GDPR compliance: data is processed in EU servers, critical for finance, healthcare, and government sectors. Additionally, Mistral offers open-source models for local deployment, giving full infrastructure control. In this article, we'll cover how we integrate Mistral Large for business tasks, Mistral Small for lightweight scenarios, and Codestral for code generation with Fill-in-the-Middle (FIM) support. A typical use case: a chatbot with document summarization in Russian, where p99 latency must not exceed 2 seconds — Mistral Large handles this at a token cost 3 times lower than GPT-4. Token savings at 1 million tokens per day amount to about $150 per month. When processing 500,000 requests per month, savings reach $4,000.

Our team has 7 years of LLM integration experience, with 50+ projects in finance and healthcare. We configure API keys, rate limiting, and monitoring to ensure compliance. Based on our measurements, p99 latency with streaming is 1.8 seconds for a 10K token context — acceptable for real-time chats.

How Mistral Large ensures GDPR compliance?

When handling confidential data, legal risks are eliminated. Mistral Large uses data centers within the EU, so data never leaves the jurisdiction.

When is Mistral Small more cost-effective than GPT-4?

For simple tasks — classification, summarization, entity extraction — Mistral Small delivers comparable quality at 5 times lower cost on input tokens. For high-volume scenarios (e.g., 100,000 requests per day), savings can reach $8,000 per month. The model also supports multilingualism, including Russian.

Mistral Large offers substantial savings — up to 5 times on input tokens. Moreover, quality on Russian is on par with GPT-4 due to multilingual pre-training.

How we set up the integration: process

We follow a standard workflow:

1. Analysis. Gather requirements: which models, request volume, latency SLA (p95 no more than 2 seconds), token budget.
2. Design. Choose connection method: official Mistral AI SDK, LangChain/LlamaIndex for RAG, or direct HTTP API. Define fallback logic for API unavailability.
3. Implementation. Write integration code — from simple chat completion to complex pipelines with function calling and streaming. All solutions are covered by tests.
4. Testing. Run load tests with different models, measure p99 latency, verify response correctness.
5. Deployment and monitoring. Deploy in your environment (AWS, GCP, on-prem), set up alerts for errors and budget overruns.

Integration via official SDK

from mistralai import Mistral
import asyncio

client = Mistral(api_key="MISTRAL_API_KEY")

# Basic call
response = client.chat.complete(
    model="mistral-large-latest",
    messages=[{"role": "user", "content": "Hello"}],
    temperature=0.1,
)
print(response.choices[0].message.content)

# Async
async def async_chat(prompt: str) -> str:
    response = await client.chat.complete_async(
        model="mistral-small-latest",
        messages=[{"role": "user", "content": prompt}],
    )
    return response.choices[0].message.content

# Streaming
with client.chat.stream(
    model="mistral-large-latest",
    messages=[{"role": "user", "content": "Long response"}],
) as stream:
    for event in stream:
        print(event.data.choices[0].delta.content or "", end="")

Function Calling

tools = [{
    "type": "function",
    "function": {
        "name": "search_db",
        "description": "Search company database",
        "parameters": {
            "type": "object",
            "properties": {
                "query": {"type": "string"},
                "limit": {"type": "integer", "default": 10},
            },
            "required": ["query"]
        }
    }
}]

response = client.chat.complete(
    model="mistral-large-latest",
    messages=[{"role": "user", "content": "Find information about client Ivanov"}],
    tools=tools,
    tool_choice="auto",
)

Codestral for code (FIM) and generation

# Fill-in-the-Middle
fim_response = client.fim.complete(
    model="codestral-latest",
    prompt="def calculate_discount(price: float,",
    suffix=") -> float:\n    return discounted_price",
    temperature=0,
    max_tokens=256,
)
print(fim_response.choices[0].message.content)

# Code generation
code_response = client.chat.complete(
    model="codestral-latest",
    messages=[{"role": "user", "content": "Write a Python function to parse CSV with error handling"}],
)

Local deployment via Ollama

ollama pull mistral:7b
ollama pull codestral:22b

from openai import OpenAI
local_client = OpenAI(base_url="http://localhost:11434/v1", api_key="ollama")
response = local_client.chat.completions.create(
    model="mistral:7b",
    messages=[{"role": "user", "content": "Hello"}],
)

With local deployment, you have full control over latency and cost — you only pay for hardware. We help choose the optimal GPU configuration (e.g., RTX 4090 for Mistral 7B or A100 for Codestral 22B).

What's included in the result

Upon completion, you receive:

• Working integration code with the chosen model (Mistral Large/Small/Codestral) supporting streaming and function calling.
• API documentation and operation instructions.
• Configured monitoring (logs, latency and error rate metrics).
• Training for your developers: 2-3 hours.
• Code warranty — 30 days free support after delivery.

Timelines: basic integration — from 1 day, complex pipeline with RAG and local deployment — up to 1 week.

Evaluate Mistral's capabilities for your project. Contact us — we'll conduct an audit within 1 day. Order a turnkey integration with quality guarantee.

LLM Development: Fine-Tuning, RAG, Agents, and Production Deployment

Using GPT‑4 or Claude 3.5 Sonnet through a public API is not a solution — it's just a tool. When the requirement is to "make it like ChatGPT, but on our data," there is a real engineering challenge behind it: from prompt engineering to training a 70B model on your own infrastructure. End-to-end LLM solution development is a complex stack, and we have been doing it for over 5 years. During this time, we have completed over 20 projects in generative AI: from RAG systems for legal departments to custom support agents. Where exactly your task falls depends on data, latency requirements, budget, and how critical confidentiality is.

A typical situation: the client has already tried ChatGPT, but results are unstable — sometimes accurate, sometimes hallucinating. Or they need integration into a corporate portal while complying with security policies. Let's break down each layer of the stack in detail — from RAG to production deployment.

Why Do RAG Systems Break and How to Fix It?

RAG (Retrieval-Augmented Generation) looks simple: find relevant documents, put them in context, get an answer. In practice, it fails in several places.

Chunking without overlap. Classic mistake: chunk_size=512, overlap=0. If the answer lies across two chunks, retrieval won't find either with sufficient confidence. Solution: overlap 15–25% of chunk_size, or better yet, sentence-aware splitting with spaCy or NLTK instead of naive character splitting.

Poor embedder. text-embedding-ada-002 is good for general use, but on legal or medical texts, specialized models like E5-large-v2, BGE-M3, or fine-tuned sentence-transformers on domain data outperform it. Recall@5 differences can be 15–25%.

No re-ranking. Vector search optimizes for speed, not relevance. A cross-encoder re-ranker (ms-marco-MiniLM-L-6-v2, bge-reranker-large) after initial retrieval improves top-3 accuracy with acceptable latency (+50–150ms). This is often more impactful than improving the embedding model.

Hybrid search. Dense vectors alone work poorly on exact queries: names, SKUs, codes. BM25 (sparse) finds exact matches but misses semantics. Hybrid via RRF (Reciprocal Rank Fusion) is the optimal compromise. Qdrant, Weaviate, and pgvector 0.7+ support hybrid search natively.

When to Fine-Tune Instead of Prompt Engineering?

Prompt engineering solves ~70% of LLM adaptation tasks for a domain. The remaining 30% require fine-tuning. Three indicators: the model ignores a specific output format even with detailed prompting; the task requires deep knowledge of specialized vocabulary (medicine, law); you need to significantly reduce token costs by replacing a large model with a smaller specialized one.

LoRA and QLoRA are the standard for SFT. LoRA adds trainable low-rank matrices to attention layers. A typical configuration for Llama-3 8B: r=64, lora_alpha=128, target_modules=["q_proj","v_proj","k_proj","o_proj"] yields ~0.8% trainable parameters, training on one A100 40GB. QLoRA adds 4-bit quantization (NF4) and allows fine-tuning 70B models on two A100 40GB, though speed drops by half compared to bf16.

DPO instead of RLHF. Direct Preference Optimization requires only (chosen, rejected) pairs, not scalar reward signals. DPOTrainer from the trl library (Hugging Face) implements it in a few dozen lines.

Common mistake. A dataset of 500 examples, 5 epochs, validation loss 0.8 — seems fine. But on test, the model degrades on general instructions. Cause: catastrophic forgetting. Solution: add 10–20% general instruction-following examples (Alpaca, FLAN) to the training set to preserve original capabilities.

How to Choose a Base Model: 8B or 70B?

For most tasks, fine-tuning an 8B model is sufficient. 70B is needed when deep reasoning is required or the 8B baseline does not reach the required quality even after fine-tuning. Inference cost for Llama-3 8B via vLLM on A100 is efficient; the exact cost depends on volume.

What Does PagedAttention Bring to Production?

vLLM is the first choice for serving open-source models. PagedAttention is the key technical innovation: KV-cache is managed like virtual memory in an OS, without fragmentation. This yields 2–4x higher throughput compared to naive HuggingFace Transformers inference. The vLLM documentation confirms that continuous batching and PagedAttention are the standard for high-load LLM services.

Typical numbers on A100 80GB for Llama-3 8B (bf16): 400–600 req/s, P50 latency 200–400ms, P99 latency 600–900ms at concurrency 64. For 70B on two A100 with tensor parallelism: 80–120 req/s, P99 latency 1.5–2.5s. AWQ or GPTQ quantization reduces memory consumption by 2x with quality loss within 1–3%.

Multi-Agent Systems

Agents are LLMs with access to tools: search, code execution, API calls, database interaction. Common patterns:

• ReAct (Reason + Act): the model reasons → chooses a tool → observes the result → reasons again. LangChain and LlamaIndex implement it out of the box.
• Multi-agent orchestration: multiple specialized agents with a coordinator on top. Example: coordinator → researcher (search + summarization) → coder (code generation and execution) → critic (verification). Tools: AutoGen (Microsoft), CrewAI, custom implementation on LangGraph.

In production, agent systems are non-deterministic. Essential: guardrails, step limits, logging of each step, human-in-the-loop for critical actions.

How We Work: Stages, Timeline, Deliverables

What Is Included

We deliver:

• Technical documentation (model card, configs, deployment instructions)
• Access to infrastructure (code repository, trained weights)
• 1 month of post-deployment support (consultations, bug fixes)
• Customer team training (2–3 sessions on system operation)

Timeline: basic RAG prototype — 1–2 weeks. Fine-tuning with customer data — 3–6 weeks (including data preparation). Production system with monitoring and retraining — 2–4 months. Cost is calculated individually based on data volume, model complexity, and infrastructure requirements.

We guarantee the quality of the final model with performance benchmarks and ongoing monitoring. Our engineers have hands‑on experience with dozens of production LLM systems.

Want to evaluate your project? Leave a request — we will prepare a preliminary summary within 1–2 business days. Or get a consultation on choosing the approach: RAG, fine-tuning, or hybrid — we will tell you what works best for you. Contact us to discuss your LLM development needs. Schedule a free consultation today.