Integrate OpenAI Assistants API for AI Agents

Customers complaining about slow support responses? Handmade RAG takes weeks to develop and constant monitoring? OpenAI Assistants API is a ready-made engine for AI agents that we deploy in days. Our team has over 5 years of AI/ML experience and 15+ successful implementations: HR bots, corporate FAQ, and tech support. With Assistants API, we cut prototype launch time by 5–10x compared to custom RAG.

How Assistants API works under the hood

Unlike Chat Completions API, Assistants handle memory and lifecycle management. Each conversation lives in a Thread—a persistent entity that stores full history. This allows the agent to retain context even after a pause of several hours. A Thread can hold up to 2000 messages (per OpenAI Assistants API), sufficient for long consultations.

Vector Store is another key feature: you upload documents (PDF, Word, CSV), and the service automatically splits them into chunks and indexes. No more debates about chunking or embedding allocation. The File Search tool returns relevant pieces directly into the assistant's prompt. Code Interpreter executes Python in an isolated sandbox—ideal for CSV analysis or calculations.

Basic integration

from openai import OpenAI
import time

client = OpenAI()

# Create an assistant
assistant = client.beta.assistants.create(
    name="Corporate Support Assistant",
    instructions="""You are a tech support assistant for TechCorp.
Help customers solve problems. Use uploaded documents as the source of truth.
Perform code for calculations and data analysis when needed.""",
    model="gpt-4o",
    tools=[
        {"type": "file_search"},
        {"type": "code_interpreter"},
    ],
)

# Create a Thread (stores history)
thread = client.beta.threads.create()

# Add a message
message = client.beta.threads.messages.create(
    thread_id=thread.id,
    role="user",
    content="How do I configure SSO for a corporate account?",
)

# Run the assistant
run = client.beta.threads.runs.create(
    thread_id=thread.id,
    assistant_id=assistant.id,
)

# Wait for completion
while run.status in ["queued", "in_progress"]:
    run = client.beta.threads.runs.retrieve(thread_id=thread.id, run_id=run.id)
    time.sleep(0.5)

# Get the response
messages = client.beta.threads.messages.list(thread_id=thread.id)
print(messages.data[0].content[0].text.value)

File Search: upload and index documents

# Create Vector Store (RAG storage)
vector_store = client.beta.vector_stores.create(name="Support Docs")

# Upload documents
with open("docs/user-guide.pdf", "rb") as f:
    file = client.beta.vector_stores.files.upload_and_poll(
        vector_store_id=vector_store.id,
        file=f,
    )

# Attach to assistant
assistant = client.beta.assistants.update(
    assistant_id=assistant.id,
    tool_resources={"file_search": {"vector_store_ids": [vector_store.id]}},
)

Custom functions + streaming

Handling function calls in real-time is what makes an assistant truly useful. Here's how it looks with streaming:

import json

# Assistant with custom functions
assistant_with_tools = client.beta.assistants.create(
    name="CRM Assistant",
    model="gpt-4o",
    tools=[{
        "type": "function",
        "function": {
            "name": "get_customer_info",
            "description": "Get customer information from CRM",
            "parameters": {
                "type": "object",
                "properties": {
                    "customer_id": {"type": "string"},
                },
                "required": ["customer_id"],
            },
        },
    }],
)

# Handle function calls with streaming
def handle_run_with_functions(thread_id: str, assistant_id: str):
    with client.beta.threads.runs.stream(
        thread_id=thread_id,
        assistant_id=assistant_id,
    ) as stream:
        for event in stream:
            if event.event == "thread.run.requires_action":
                # Execute functions
                tool_outputs = []
                for tool_call in event.data.required_action.submit_tool_outputs.tool_calls:
                    if tool_call.function.name == "get_customer_info":
                        args = json.loads(tool_call.function.arguments)
                        result = crm.get_customer(args["customer_id"])
                        tool_outputs.append({
                            "tool_call_id": tool_call.id,
                            "output": json.dumps(result),
                        })

                # Submit results
                stream.submit_tool_outputs_and_stream(tool_outputs)

            elif event.event == "thread.message.delta":
                for delta in event.data.delta.content:
                    if delta.type == "text":
                        print(delta.text.value, end="", flush=True)

Why is this faster than custom RAG?

Assistants API wins on launch speed: a basic prototype is ready in 1–3 days versus 2–4 weeks for a custom pipeline. However, it sacrifices flexibility: no hybrid search, limited prompt control. For production with high search quality demands, we build a custom pipeline on LangChain.

Practical case: corporate FAQ assistant for HR

From our practice: a large retailer received 50+ repetitive questions from employees daily (vacations, documents, benefits). The HR manager spent 2 hours each day on standard replies. We built an agent on Assistants API with File Search: uploaded 15 policy documents to Vector Store and integrated with corporate Slack.

Results:

• Autonomous responses to 73% of questions (saving ~$1,200 per month in HR salary);
• Deployment time: 5 days (vs. 2 weeks for custom RAG);
• HR manager gained 1.5 free hours per day.

Limitations encountered: high Vector Store storage costs and inability to fine-tune search. For the next project, we recommended a custom stack—but in this case, speed of launch outweighed.

What's included in the work

We don't just plug in an API—we design the agent for your infrastructure. The work process includes the following stages:

1. Scenario analysis and architecture design (Threads, tools, integrations);
2. Vector Store setup—upload and index your knowledge base;
3. Custom function implementation (CRM, 1C, ERP);
4. Streaming setup and function call handling;
5. Integration with corporate messengers (Slack, Telegram, Teams);
6. Documentation and team training;
7. Post-launch support (SLA 99.9% guaranteed).

Timeline and cost

• Basic assistant + File Search: from 1 to 3 days
• Custom functions + streaming: from 3 to 5 days
• Production deployment with monitoring, logging, and CI/CD: from 1 week

Cost is calculated individually—based on the number of scenarios, document volume, and integration complexity. We provide a transparent estimate before starting work. Our AI solutions are already used by over 20 companies—contact us to discuss your project and get a consultation to evaluate it under your budget.

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.