Anthropic Claude Integration: Tool Use, Caching, and Optimization

Anthropic Claude Integration: Tool Use, Caching, and Optimization

A client approached us to integrate Claude into their SaaS product. The first version called the model with no optimization at all. After a week of testing, the API bill hit thousands of dollars, and the response quality didn't meet expectations. We redesigned the architecture: implemented Tool Use for agentic scenarios, configured Prompt Caching for static instructions, and selected the right model for each query type. The result—cost reduced by 80% while maintaining p99 latency under 2 seconds. This article covers practical steps useful for anyone deploying Claude into production.

"The integration reduced our costs by 80% while improving response times." — Client

Which Claude Model to Choose for Production?

Claude Haiku is 5x cheaper than Opus with similar quality on simple tasks. The choice depends on your scenario; we help select the optimal configuration for your load.

Why Model Selection Determines Your Budget

On one project, we replaced Opus with Sonnet for 70% of requests, keeping Opus only for complex reasoning. This cut overall API costs by 3x. For simple tasks like classification or data extraction, Haiku provides the same accuracy as Opus but costs 10x less. Always test on your own data.

How to Set Up Basic Integration

import anthropic
from pydantic import BaseModel

client = anthropic.Anthropic()  # ANTHROPIC_API_KEY from env

# Basic call
def chat(prompt: str, model: str = "claude-sonnet-4-5") -> str:
    message = client.messages.create(
        model=model,
        max_tokens=1024,
        messages=[{"role": "user", "content": prompt}]
    )
    return message.content[0].text

# With system prompt
def chat_with_system(system: str, prompt: str) -> str:
    message = client.messages.create(
        model="claude-sonnet-4-5",
        max_tokens=2048,
        system=system,
        messages=[{"role": "user", "content": prompt}],
        temperature=0.1,
    )
    return message.content[0].text

# Streaming
def stream_response(prompt: str):
    with client.messages.stream(
        model="claude-sonnet-4-5",
        max_tokens=1024,
        messages=[{"role": "user", "content": prompt}],
    ) as stream:
        for text in stream.text_stream:
            yield text

# Vision
def analyze_image(image_base64: str, media_type: str, question: str) -> str:
    message = client.messages.create(
        model="claude-sonnet-4-5",
        max_tokens=1024,
        messages=[{
            "role": "user",
            "content": [
                {
                    "type": "image",
                    "source": {
                        "type": "base64",
                        "media_type": media_type,
                        "data": image_base64,
                    },
                },
                {"type": "text", "text": question},
            ],
        }]
    )
    return message.content[0].text

What Is Tool Use and How to Build an Agent?

Tool Use (Function Calling) allows Claude to call external functions. You describe tools in JSON Schema, and the model decides when to invoke them. This is the core mechanism for building agents.

tools = [{
    "name": "get_weather",
    "description": "Get current weather for a city",
    "input_schema": {
        "type": "object",
        "properties": {
            "city": {"type": "string", "description": "City name"},
            "units": {"type": "string", "enum": ["celsius", "fahrenheit"]},
        },
        "required": ["city"]
    }
}]

def run_agent_loop(user_message: str) -> str:
    messages = [{"role": "user", "content": user_message}]

    while True:
        response = client.messages.create(
            model="claude-sonnet-4-5",
            max_tokens=1024,
            tools=tools,
            messages=messages,
        )

        if response.stop_reason == "end_turn":
            return response.content[-1].text

        # Handle tool_use blocks
        tool_results = []
        for block in response.content:
            if block.type == "tool_use":
                result = dispatch_tool(block.name, block.input)
                tool_results.append({
                    "type": "tool_result",
                    "tool_use_id": block.id,
                    "content": str(result),
                })

        messages.append({"role": "assistant", "content": response.content})
        messages.append({"role": "user", "content": tool_results})

With multiple tools, the logic is the same—the model chooses which to call. Proper error handling is important: pass errors back with is_error in the next request.

How Does Prompt Caching Work and Why Can It Save Up to 90%?

Caching large static prompts (documents, instructions) is a key technique. The cache is stored for 5 minutes and savings on repeated calls can reach 90%. Anthropic recommends caching blocks over 1024 tokens.

def cached_analysis(system_doc: str, question: str) -> str:
    message = client.messages.create(
        model="claude-sonnet-4-5",
        max_tokens=1024,
        system=[{
            "type": "text",
            "text": system_doc,
            "cache_control": {"type": "ephemeral"},  # Cache this block
        }],
        messages=[{"role": "user", "content": question}]
    )
    return message.content[0].text

In practice: if you frequently repeat the same instructions (e.g., response format description), move them into a separate cacheable block. This reduces latency and cost.

Comparison: With and Without Caching

Savings on repeated instructions can reach 90%. After optimization, our clients typically save between 50% and 90% on API costs.

Example savings calculation for 10,000 requests per day: Assume each request contains a static instruction of 500 tokens. Without caching, you pay for all tokens. With caching, the instruction is charged only on its first occurrence within a 5-minute window. With evenly distributed requests, the cache hits for 80% of requests, reducing costs for the same number of tokens. On Sonnet, this saves roughly $200 per month.

What Our Integration Includes

• Architectural documentation with integration schema and model selection
• Code repository in Python (FastAPI or Flask) with streaming, Vision, and Tool Use support
• Monitoring and alerting setup (Grafana + Prometheus) for key metrics: p99 latency, tokens per minute, errors
• Customer team training (2–3 hours) on operation and further development
• Support for 30 days after deployment

With over 10 years of experience and more than 50 deployed AI integrations, we have the expertise to handle your project.

Step-by-Step Integration Guide

1. Analytics — define the scenario, load, and choose the model.
2. Design — integration architecture, rate limiting, and retry logic.
3. Implementation — basic calls, streaming, Vision.
4. Tool enablement — Tool Use for agents.
5. Optimization — Prompt Caching, model selection, temperature tuning.
6. Testing — performance under load, measurement of p99 latency.
7. Deployment — roll out, monitoring, team training.

Each step can be done separately. A full cycle takes up to a week.

Timelines and Cost

• Basic integration: from 0.5 days
• Tool Use + agent loop: 2–3 days
• Prompt Caching + optimization: 1 day
• Full cycle: up to a week

Cost is calculated individually based on complexity. Our basic integration starts at $500, with full turnkey solutions from $2,000. Write to us to get a free estimate for your project.

Typical Integration Mistakes

• Ignoring caching — leads to unnecessary costs (could have saved 50–90%).
• Wrong model selection — Haiku suffices for 70% of requests, yet Opus is used.
• No retry logic — requests are lost due to rate limits.
• Overly long system prompts without caching — increases latency and cost.

We provide turnkey integration in as little as 2 days. Contact us to discuss your integration scenario. We guarantee stable operation and cost optimization.

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.