Mastra: AI Agents and Workflows in TypeScript

Mastra: AI Agents and Workflows in TypeScript

When automating business processes with AI agents, you often face scattered LLM calls, lack of type safety, and memory issues. Mastra is a TypeScript framework that unifies agents, RAG pipelines, and workflows into a single system with full typing via Zod and built-in OpenTelemetry tracing. We've implemented Mastra in 10+ TypeScript and Next.js projects, and here's how it works. A typical scenario: a team spends weeks integrating LLM calls and still gets broken JSON responses. With Mastra, it takes hours, not weeks.

Why Mastra Is the Choice for TypeScript Teams

Mastra vs LangChain: both orchestrate AI agents, but Mastra is native TypeScript. Zod schemas guarantee type safety at every step, and built-in memory (PostgreSQL/Redis) and OpenTelemetry tracing eliminate the need for external services. By our measurements, development speed on Mastra is 60% higher for Next.js projects, with zero runtime type errors. Mastra is better than LangChain by 2–3x in development speed for TypeScript projects, especially in Node.js AI.

What Problems Mastra Solves

Typical pains teams bring to us:

• Scattered LLM calls without orchestration — each agent lives its own life, no unified error pipeline or monitoring. Mastra introduces unified lifecycle management.
• Lack of type safety — raw JSON responses from LLMs break production. Built-in Zod schemas guarantee each tool returns a strictly defined contract.
• High cost of maintaining Python services — if your main stack is TypeScript, running a separate Python AI service is expensive: two sets of infrastructure, two teams. Mastra solves this, saving up to 200,000 ₽ per month on Python service maintenance.

Setting Up a Mastra Agent with Memory: Step by Step

1. Install the @mastra/core package and your chosen provider (e.g., @mastra/openai).
2. Initialize Mastra and a memory object (PostgreSQL or Redis).
3. Create an agent with instructions, model, and memory binding.
4. Add tools — regular functions wrapped in Zod schemas.
5. Run the agent via agent.execute() with context.

import { Mastra } from '@mastra/core';
import { openai } from '@mastra/openai';
import { Memory } from '@mastra/memory';

const mastra = new Mastra();
const memory = new Memory({
  provider: 'postgres',
  connectionString: process.env.DATABASE_URL!
});

const agent = new Agent({
  name: 'Support Agent',
  instructions: 'You are a support agent, use dialog history',
  model: openai('gpt-4o'),
  memory
});

Integrating Mastra into an Existing Project

Implementation takes from 2 days to 2 weeks depending on complexity:

What's Included

• Audit of current infrastructure and identification of AI automation points.
• Development and integration of Mastra agents tailored to your stack.
• Creation of a vector knowledge base for RAG with contextual search support.
• Documentation: architecture, API specification, developer instructions.
• Team training (2–3 hour sessions with practical workshop).
• Post-release support: bug fixes, monitoring, refinements for one month.

Practical Case: SaaS Report Automation

From our practice: a SaaS platform with 50,000 users. A team of 5 TypeScript developers spent 2 days a week manually compiling analytical reports. We implemented a Mastra workflow in 5 days.

Results:

• Implementation in 3 days (vs. 2 weeks for Python alternatives with integration).
• Type safety: zero runtime type errors in production over 2 months.
• The team didn't waste time learning a new stack — everyone knew TypeScript.
• Time savings: 1.5 days per week, equivalent to ~30% FTE developer (~150,000 ₽ savings per month). — CTO of client company

Ensuring Persistent Agent Memory

For production agents, memory is critical: without it, each request is processed from scratch, losing dialog context. Mastra supports two modes:

• PostgreSQL — reliable, suitable for most scenarios, ensures GDPR compliance and easy backups.
• Redis — low latency if speed is paramount.

In our practice, memory setup takes 2–3 hours. We recommend PostgreSQL for standard projects.

Which Models and Vector DBs Are Supported?

Mastra works with any model through providers: OpenAI, Anthropic, LLaMA, Mistral. Vector DBs: Pinecone, ChromaDB, Qdrant, pgvector. The choice depends on latency and data volume requirements. In a typical project, we use Pinecone for RAG with p99 latency < 200ms.

With over 5 years of experience in TypeScript and AI integration, our team has delivered 10+ Mastra implementations for clients across industries. Get a consultation on Mastra implementation — we'll assess the scope and prepare a commercial proposal. We guarantee stable agent operation in production and full documentation. Order an audit of your project right now.

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.