Photo by the author
# Entry
The rise of frameworks such as LangChain AND CrewAI has made building AI agents easier than ever. However, creating these agents often involves pushing API speed limits, managing multi-dimensional data, or exposing local servers to the Internet.
Instead of paying for cloud services during the prototyping stage or polluting your host computer with dependencies, you can take advantage Docker. With one command, you can speed up infrastructure that makes your agents smarter.
Here are 5 must-have Docker containers that every AI agent developer should have in their toolkit.
# 1. Ollama: Run local language models
Ollam’s dashboard
When creating agents, send all prompts to the cloud provider, e.g OpenAI it can become pricey and snail-paced. Sometimes you need a brisk, private model for specific tasks – such as grammar correction or classification tasks.
To be allows you to run open source vast language models (LLM) – e.g Lama 3, MistralOr Phi — directly on your local computer. By running it in a container, you keep your system tidy and can easily switch between different models without complicated Python setup.
Privacy and costs are the main issues that construction agents face. The Ollam docker image makes it simple to support models like Llama 3 or Mistral via the REST API.
// Explaining why this is significant for agent developers
Instead of sending sensitive data to external APIs like OpenAI, you can give your agent a “brain” that lives in your own infrastructure. This is significant for enterprise agents who handle proprietary data. Running docker run ollama/ollamayou immediately have a local endpoint that your agent code can call to generate text or justify tasks.
// Initiating a quick start
To download and run the Mistral model via an Ollama container, employ the following command. This will map the port and save the models on your local drive.
docker run -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollamaOnce the container is running, you need to download the model by executing the command inside the container:
docker exec -it ollama ollama run mistral// Explaining why this is useful for agent developers
You can now designate your agent’s LLM client http://localhost:11434. This provides an API-compliant local endpoint for rapid prototyping and ensures that data never leaves your machine.
// Overview of key benefits
- Data Privacy: Protect your prompts and data
- Cost savings: No API fees for inference
- Latency: Faster responses when running on local GPUs
Find out more: Ollama docking hub
# 2. Qdrant: vector memory database
Qdrant panel
Agents need memory to recall past conversations and domain knowledge. To provide the agent with long-term memory, you need: vector database. These databases store numerical representations (embeddings) of text, allowing the agent to later search for semantically similar information.
Qdrant is a high performance open source vector database built in Rust. It’s brisk, reliable and offers both gRPC and the REST API. Running it in Docker provides an immediate, production-grade storage system for agents.
// Explaining why this is significant for agent developers
To build a Search Augmented Generation (RAG) agent, you need to store embedded documents and retrieve them quickly. Qdrant serves as the agent’s long-term memory. When a user asks a question, the agent converts it into a vector, searches Qdrant for similar vectors – representing relevant knowledge – and formulates an answer based on this context. Running it in Docker separates this memory layer from the application code, making it more reliable.
// Initiating a quick start
You can run Qdrant with a single command. This will expose the API and dashboard on port 6333 and the gRPC interface on port 6334.
docker run -d -p 6333:6333 -p 6334:6334 qdrant/qdrantOnce launched, you can connect to your agent localhost:6333. When the agent learns something recent, record the embedding in Qdrant. The next time a user asks a question, the agent can search this database for relevant “memories” that can be included in the question, making the conversation truly conversational.
# 3. n8n: Glue workflows together
Dashboard n8n
Agentic workflows rarely exist in a vacuum. Sometimes you need an agent to check your email, update a row in a Google Sheet, or send a message in Slack. Although API calls can be written by hand, the process is often tedious.
n8n is a workflow automation tool based on straightforward code. It allows you to connect different services using a visual user interface. By running it locally, you can create sophisticated workflows—for example, “If an agent detects a lead, add them to HubSpot and send a Slack alert”—without writing a single line of integration code.
// Initiating a quick start
To maintain your workflows, you must mount the volume. The following command configures n8n with SQLite as the database.
docker run -d --name n8n -p 5678:5678 -v n8n_data:/home/node/.n8n n8nio/n8n// Explaining why this is useful for agent developers
You can design your agent to call the n8n webhook URL. The agent simply sends the data and n8n takes care of the messy logic of communicating with third-party APIs. This separates the “brain” (LLM) from the “hands” (integrations).
Access the editor at http://localhost:5678 and start automating.
Find out more: n8n docking center
# 4. Firecrawl: Transform web pages into multilingual, model-ready data
Firecrawl panel
One of the most common tasks for agents is research. However, agents have difficulty reading raw web pages rendered in HTML or JavaScript. They need plain text in Markdown format.
Firecrall is an API service that takes a URL, crawls a site, and converts its content into pure markdown or structured data. It supports JavaScript rendering and automatically removes templates such as ads and navigation bars. Running it locally bypasses the usage restrictions of the cloud version.
// Initiating a quick start
Firecrawl uses, among others: docker-compose.yml file because it consists of multiple services including Apps, Redis and Playwright. Clone the repository and run it.
git clone https://github.com/mendableai/firecrawl.git
cd firecrawl
docker compose up// Explaining why this is useful for agent developers
Give your agent the ability to accept live internet data. If you are creating a research agent, you can ask it to invoke a local instance of Firecrawl to download a web page, convert it to plain text, chunk it, and save it autonomously to a Qdrant instance.
# 5. PostgreSQL and pgvector: Implementation of relational memory
PostgreSQL dashboard
Sometimes vector search alone is not enough. You may need a database that can simultaneously support structured data – such as user profiles or transaction logs – and vector embeddings. PostgreSQLWith pgvector extension, allows you to do just that.
Instead of running a separate vector database and a separate SQL database, you get the best of both worlds. You can store a user’s name and age in a column in the table, and the embedding of their conversations in another column, and then perform a hybrid search (e.g., “Find New York users’ conversations about refunds”).
// Initiating a quick start
The official PostgreSQL image does not include pgvector by default. You need to employ a specific image, for example the one from the pgvector organization.
docker run -d --name postgres-pgvector -p 5432:5432 -e POSTGRES_PASSWORD=mysecretpassword pgvector/pgvector:pg16// Explaining why this is useful for agent developers
This is the best resource for state agents. Your agent can save its memories and internal state in the same database as your application data, ensuring consistency and simplifying the architecture.
# Summary
You don’t need a huge cloud budget to build sophisticated AI agents. The Docker ecosystem provides production-grade alternatives that run perfectly on a developer’s laptop.
By adding these five containers to your workflow, you’ll equip yourself with:
- Brains: Ollama for local inference
- Memory: Qdrant for vector search
- Hands: n8n for workflow automation
- Eyes: Firecrawl to capture webs
- Storage: PostgreSQL with pgvector for structured data
Launch containers, point your LangChain or CrewAI code to localhost, and watch your agents come to life.
// Further reading
Shittu Olumid is a software engineer and technical writer with a passion for using cutting-edge technology to create compelling narratives, with an eye for detail and a knack for simplifying sophisticated concepts. You can also find Shittu on Twitter.