MetaGPT Integration for Multi-Agent Software Development

Imagine: your team spends 20+ hours writing a standard CRUD module — model, repository, service, controller, tests. We integrate MetaGPT — a multi-agent framework that automates this routine, freeing developers for complex business tasks. At its core is a simulation of a full-stack team: Product Manager generates PRD, Architect designs the system, Engineer writes code, QA creates tests. MetaGPT uses SOP (Standard Operating Procedures) as the coordination foundation, ensuring consistency and quality of artifacts. Our experience: over 5 years in AI development, 50+ projects with LLMs, certified engineers.

A typical CRUD module is generated in 40 minutes instead of 5 hours — a 7x difference. Test coverage increases from 73% to 89%, and code becomes uniform. We guarantee integration into your current stack within 1–2 weeks. Request a consultation — our engineers will analyze your process and propose the optimal implementation.

How MetaGPT Generates Code

The process starts with describing an idea in natural language. Below is the basic workflow we adapt to the project:

1. Install the package: pip install metagpt
2. Configure the API key of the chosen LLM (GPT-4, Claude, LLaMA)
3. Describe the task as a string in natural language
4. Run asynchronous generation with parameters (investment, n_round, code_review)
5. Get the repository structure with code, tests, and documentation

# pip install metagpt
import asyncio
from metagpt.software_company import generate_repo, ProjectRepo

async def develop_feature(requirement: str) -> ProjectRepo:
    repo = await generate_repo(
        idea=requirement,
        investment=3.0,
        n_round=5,
        code_review=True,
        run_tests=True,
    )
    return repo


result = asyncio.run(develop_feature(
    "Разработай REST API для управления задачами: CRUD операции, приоритеты, назначение на исполнителей. Python + FastAPI + PostgreSQL."
))

print(result.get_code())
print(result.get_tests())
print(result.get_docs())

Custom Roles for Your Process

Standard roles are often insufficient. We add, for example, Security Engineer for security auditing. The role inherits from Role and defines its own actions.

from metagpt.roles import Role
from metagpt.actions import Action
from metagpt.schema import Message

class SecurityAuditAction(Action):
    name: str = "SecurityAudit"
    i_context: str = ""

    async def run(self, code: str) -> str:
        prompt = f"""Проведи аудит безопасности следующего кода.
Проверь: SQL injection, XSS, hardcoded secrets, небезопасные зависимости,
отсутствие валидации входных данных.

Код:
{code}

Формат: список уязвимостей с severity (Critical/High/Medium/Low) и рекомендациями."""
        return await self._aask(prompt)


class SecurityEngineer(Role):
    name: str = "Security Engineer"
    profile: str = "Security Engineer"
    goal: str = "Обеспечить безопасность кода перед деплоем"
    constraints: str = "Проверяй только код, написанный командой"

    def __init__(self, **kwargs):
        super().__init__(**kwargs)
        self._init_actions([SecurityAuditAction])
        self._watch([WriteCode])

    async def _act(self) -> Message:
        todo = self.rc.todo
        code = self.get_memories(k=1)[0].content
        audit_result = await todo.run(code)
        return Message(
            content=audit_result,
            role=self.profile,
            cause_by=type(todo),
        )

We can also create roles for specific tasks — DevOps Engineer for infrastructure generation, DBA for query optimization. Each role inherits from Role and defines its own actions.

from metagpt.config import Config
config = Config(
    llm={
        "api_type": "openai",
        "model": "gpt-4",
        "api_key": "sk-...",
    },
    role={
        "name": "DevOps Engineer",
        "profile": "DevOps Engineer",
        "goal": "Генерация Dockerfile и CI/CD конфигов",
        "constraints": "Используй последние версии инструментов",
    }
)

When MetaGPT Is Inefficient?

MetaGPT handles structural code well: CRUD, REST API, data models. But complex business logic with non-trivial algorithms requires significant refinement. For example, if the condition has >3 levels of nesting or requires an original algorithm — it's better to write manually. We always determine the applicability boundaries before implementation.

Practical Case: 7x Acceleration of CRUD Generation

Problem: developers spent 4–6 hours on a standard CRUD module (model, repository, service, controller, tests). 20+ such modules per quarter.

Approach: MetaGPT generates boilerplate code from a specification, the developer reviews and refines business logic.

Specification template:

spec_template = """
Создай CRUD-модуль для сущности {entity_name}:
- Поля: {fields}
- Отношения: {relations}
- Бизнес-правила: {rules}
- Стек: FastAPI + SQLAlchemy 2.0 + Alembic + pytest
- Включи: pydantic схемы, репозиторий, сервис, роутер, тесты CRUD
"""

Developers appreciated the reduction in routine but noted the need for business logic review.

Why Implement MetaGPT?

The main reason is speed. CRUD generation accelerates by 7x, test coverage increases by 16%, and code becomes uniform. The team stops drowning in routine and focuses on unique business logic. An additional effect is reducing errors through automated test generation and agent code review. Our engineers with 10+ years of experience in AI/ML guarantee implementation quality.

What Our Work on MetaGPT Integration Includes

• Audit of the current development process: identify templated tasks suitable for automation.
• MetaGPT configuration for your stack: model configuration (GPT-4, Claude, LLaMA), vector database selection (ChromaDB, pgvector), prompt fine-tuning.
• Creation of custom roles: Security Engineer, DevOps Engineer, and others for your processes.
• CI/CD integration: GitHub Actions, GitLab CI — auto-build PRs from issues.
• Documentation and team training: 2–3 workshops, review guides.
• Post-release support: 2 weeks of maintenance, adjustments based on feedback.

CI/CD Integration

We connect MetaGPT to your pipeline: when an issue with the label metagpt-generate is created, code is automatically generated and a PR is created. The team reviews and merges. This reduces time-to-market for templated tasks.

Timeline

• Basic prototype: 1–2 days.
• Custom roles and processes: 1–2 weeks.
• Full integration with CI/CD and training: 2–3 weeks.
• Cost is calculated individually after audit.

MetaGPT: open-source multi-agent framework for automating software development. Official documentation: https://github.com/geekan/MetaGPT

Get a consultation — our engineers with 10 years of experience in AI/ML will help implement MetaGPT with a guaranteed result.

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.