<p class="content-editor__unstyled _slug__documentArticle__unstyled__wrYfJ"><strong style="font-weight:bold">Docker Compose</strong> is a way to define and run several containers together as one multi-service application.</p><p class="content-editor__unstyled _slug__documentArticle__unstyled__wrYfJ">Instead of starting containers one by one with long commands, teams can describe services, ports, volumes, and dependencies in <u style="text-decoration:underline">one Compose file</u> and run the system more consistently.</p><p class="content-editor__blockquote _slug__documentArticle__blockquote__n61Jq"><span style="font-size:1.7rem">Compose is often the first place where teams stop thinking in single containers and start thinking in <a class="draft-link" href="/explains/what-is-a-stack" target="_blank" rel="nofollow">stacks</a>.</span></p><h2 class="content-editor__header-two _slug__documentArticle__header-two__YLtFS" id="why-teams-use-docker-compose" data-text-content="Why teams use Docker Compose">Why teams use Docker Compose</h2><ul><li class="content-editor__unordered-list-item _slug__documentArticle__unordered-list-item__oWKA6">to run several services together locally</li><li class="content-editor__unordered-list-item _slug__documentArticle__unordered-list-item__oWKA6">to keep service wiring more explicit</li><li class="content-editor__unordered-list-item _slug__documentArticle__unordered-list-item__oWKA6">to reproduce a setup more easily on another machine</li><li class="content-editor__unordered-list-item _slug__documentArticle__unordered-list-item__oWKA6">to avoid rebuilding the whole environment from memory each time</li></ul><div class="content-editor__header-two _slug__documentArticle__header-two__YLtFS"><br/></div><h2 class="content-editor__header-two _slug__documentArticle__header-two__YLtFS" id="why-it-matters-in-trydirect-and-stacker" data-text-content="Why it matters in TryDirect and Stacker">Why it matters in TryDirect and Stacker</h2><p class="content-editor__unstyled _slug__documentArticle__unstyled__wrYfJ"><a class="draft-link" href="/explains/what-is-stacker" target="_self">Stacker</a> generates and works with Compose-style artifacts as part of a broader stack workflow. That is one reason Docker Compose remains such a useful mental model for users moving from local experiments into repeatable deployment.</p>

Docker Compose is a way to define and run several containers together as one multi-service application.
Instead of starting containers one by one with long commands, teams can describe services, ports, volumes, and dependencies in one Compose file and run the system more consistently.
Compose is often the first place where teams stop thinking in single containers and start thinking in stacks.
Why teams use Docker Compose
to run several services together locally
to keep service wiring more explicit
to reproduce a setup more easily on another machine
to avoid rebuilding the whole environment from memory each time

Why it matters in TryDirect and Stacker
Stacker generates and works with Compose-style artifacts as part of a broader stack workflow. That is one reason Docker Compose remains such a useful mental model for users moving from local experiments into repeatable deployment.

What Docker Compose is, why teams use it for multi-service systems, and how it connects to stack-based workflows in TryDirect and Stacker.

A practical explanation of Docker Compose and why it is so important for running and understanding multi-service self-hosted stacks.

What Is Docker Compose?

n8n is a workflow automation tool that helps teams connect apps, APIs, and business steps into repeatable flows.

In practice, many teams use n8n to watch for an event, transform data, call an API or model, and push the result into another system.

If you are exploring AI and automation on TryDirect, n8n often becomes one of the most useful building blocks in the stack.

glue logic between internal systems and external services

n8n is rarely deployed as a standalone idea. It usually runs alongside a database, secrets, optional AI services, and other supporting components, which is why it is best understood as part of a stack rather than a lone app.

whether linked APIs or AI providers are still reachable

That is also why n8n appears so often in the AI experiments and operations articles: once workflows matter to the business, teams need more than a first install.

What n8n is, what it is used for, and why it is such a common component in automation and AI stacks on TryDirect.

A practical explanation of n8n and why it is such a common building block in automation and AI stacks on TryDirect.

What Is n8n?

Automation

One of the first AI stack decisions teams face is whether to start with local models, hosted providers, or a mix of both.

The right answer depends less on hype and more on what the team is optimizing for: privacy, speed of evaluation, model quality, cost control, or operational simplicity.

A good default rule is simple: choose the path that helps you learn fastest without creating unnecessary operational pain.

to test workflows without sending data to external APIs

to use tools such as Ollama for local AI paths

to explore the stack before committing to provider cost

to avoid operating local inference at the start

to compare business value before optimizing infrastructure

to move faster when local hardware is not the priority

A lot of teams eventually want both: local paths for privacy or cost-sensitive work, and hosted paths for stronger models or faster evaluation. That is one reason gateway layers such as LiteLLM become useful later.

The AI experiments article shows this decision in a broader operational context.

How to choose between local models and hosted providers based on privacy, evaluation speed, quality, cost, and operational tradeoffs.

A practical guide to choosing between local models, hosted providers, or a mix of both in self-hosted AI stacks.

How to Choose Between Local Models and Hosted Providers

Vector search is a way of finding related information by semantic similarity instead of only exact keyword matching.

In practical AI systems, vector search helps the stack find the most relevant pieces of data so an application can retrieve better context before answering or acting.

If you hear teams talk about embeddings, retrieval, or memory in AI workflows, vector search is often part of what makes that possible.

to search documents by meaning instead of only keywords

to support RAG-style assistants and knowledge systems

to help tools such as OpenClaw retrieve better context

to make AI workflows more useful with business data

Vector search usually appears through tools such as Qdrant and becomes one more important supporting layer inside an AI stack.

What vector search is, why it matters in AI systems, and how it connects to retrieval, Qdrant, and real business data.

A practical explanation of vector search and why it matters in self-hosted AI systems that work with real documents and business data.

What Is Vector Search?

Open WebUI is a user-facing interface layer for interacting with AI models and related workflows.

Teams often use it when they want a cleaner chat-style or operator-facing UI on top of local or hosted model infrastructure.

It is best understood as the visible front door to an AI stack, not as the whole stack itself.

supporting services and app logic around the user experience

sometimes other AI apps or workflows that need a clearer human interface

Open WebUI appears in the same family of practical AI stacks as OpenClaw, local model runtimes, and gateway layers because teams often need both a model path and a usable interface.

What Open WebUI is, why teams use it, and how it fits into self-hosted AI stacks as the user-facing interface layer.

A practical explanation of Open WebUI and why teams use it as the interface layer for self-hosted AI stacks.

What Is Open WebUI?

Langflow is a visual flow builder for LLM applications and AI workflows.

Teams often use it to design prompt paths, model interactions, and retrieval-oriented logic through a visual canvas rather than only code.

Like Flowise, Langflow is usually one part of a broader stack, not the whole system by itself.

to iterate on prompt and chain structure faster

to test retrieval and context flows with less friction

to compare visual AI workflow approaches with tools like Flowise

Langflow belongs in the same family of practical self-hosted AI tools that become more useful when they run in a clean stack with retrieval, models, and supporting services around them.

What Langflow is, why teams use it, and how it fits into self-hosted AI stacks with models, retrieval, and supporting services.

A practical explanation of Langflow and why teams use it as a visual AI workflow layer inside self-hosted stacks.

What Is Langflow?

Flowise is a visual AI workflow builder that helps teams compose LLM and retrieval-style flows through a more graphical interface.

In practical terms, teams often use Flowise when they want to connect prompts, models, memory, retrieval, and other AI components without writing every step from scratch.

It is easiest to understand as one interface layer inside a broader stack, not as the whole platform by itself.

to prototype retrieval and agent-style flows

to give non-backend specialists a clearer interface for iteration

to pair with other tools such as n8n and vector databases

Flowise often appears in self-hosted AI stacks alongside retrieval tools and supporting services, which is why it is helpful to read it together with the RAG explainer.

What Flowise is, why teams use it, and how it fits into self-hosted AI stacks alongside retrieval and automation tools.

A practical explanation of Flowise and why it often appears as a visual AI workflow layer inside self-hosted stacks.

What Is Flowise?

LiteLLM is an LLM gateway layer. Teams use it to normalize access to different model providers behind a more consistent interface.

In practice, that means one stack can switch between providers or combine several providers without forcing every app in the system to speak a different API shape.

If you are building a multi-service AI stack, LiteLLM is often less about one more AI app and more about cleaner provider control.

to switch or compare providers more easily

to hide provider differences behind one gateway

to mix local and hosted model paths more cleanly

to support AI workflows that may evolve over time

an app such as OpenClaw or a chat interface sends requests to LiteLLM

LiteLLM forwards requests to one or more model providers

the rest of the stack stays more stable even if the provider strategy changes

LiteLLM appears in the larger AI experiments story because it helps teams keep provider strategy flexible while the rest of the stack becomes more operationally stable.

It is especially useful when a team expects its model setup to change while keeping the broader workflow intact, including retrieval-oriented flows such as RAG.

What LiteLLM is, why teams use it as an LLM gateway, and how it helps keep provider strategy more flexible in AI stacks.

A practical explanation of LiteLLM and why teams use it as an LLM gateway layer inside multi-service AI stacks.

What Is LiteLLM?

Ollama is a local AI runtime that lets teams run models on their own machine or server instead of sending every request to a hosted provider.

In TryDirect and Stacker discussions, Ollama often appears as the simplest private path for local AI-assisted generation, testing, and experimentation.

A practical mental model: if a hosted provider is the remote AI service, Ollama is often the local AI service.

to experiment privately with local models

to avoid sending early tests to external APIs

to support local AI workflows during stack setup

to pair with tools such as OpenClaw or RAG-style systems

Stacker uses Ollama as the default local AI provider for some AI-assisted workflows, including AI-powered initialization paths.

Ollama is useful when privacy, local control, or low-friction experimentation matter more than always using a hosted commercial model.

What Ollama is, why teams use it for local AI, and how it fits into Stacker and self-hosted AI workflows.

A practical explanation of Ollama and why it is such a common local AI runtime in self-hosted experiments and Stacker workflows.

What Is Ollama?

RAG stands for retrieval-augmented generation.

In plain language, it means an AI system looks up relevant information first and then uses that retrieved context while producing an answer.

This is one of the main ways teams make AI systems more useful with real business data instead of relying only on the model’s base training.

support systems that need current business context

AI workflows that should answer from company data instead of generic memory

an application layer such as OpenClaw or a chat UI

an AI model that can answer using retrieved context

RAG is a good example of why AI work is usually a stack problem, not just a model problem. The AI experiments article shows this clearly in practice.

What RAG is, how retrieval-augmented generation works, and why it matters for self-hosted AI systems connected to real data.

A practical explanation of RAG and why retrieval matters when AI systems need to work with real business data.

What Is RAG?

Qdrant is a vector database. Teams use it to store and search embeddings so AI systems can retrieve relevant information instead of answering only from the model’s base training.

In practical terms, Qdrant often becomes the retrieval layer in a stack that supports document search, internal knowledge assistants, or AI workflows connected to company data.

If you hear people talk about retrieval, memory, or vector search in AI systems, Qdrant is one of the tools they often mean.

searching internal documents by semantic similarity

feeding relevant context into AI workflows

supporting knowledge assistants and RAG-style systems

connecting business data to AI interfaces more usefully

Qdrant often appears in the same stack as OpenClaw, n8n, or chat interfaces because those systems become much more useful when they can retrieve relevant information from real data.

Why this is a stack concern, not only a database concern

Qdrant still depends on the rest of the system around it: ingestion flows, applications, model layers, and operations. That is why it is better understood as one component inside a stack.

If you want the bigger picture, the AI experiments article shows how tools like Qdrant fit into a realistic self-hosted AI environment.

What Qdrant is, what vector search means, and why Qdrant is such a common component in self-hosted AI stacks.

A practical explanation of Qdrant and why vector search matters in self-hosted AI stacks.

What Is Qdrant?

OpenClaw is an AI application that often appears in TryDirect discussions as part of a broader self-hosted AI workflow.

The key idea is that OpenClaw is usually not the whole system by itself. Teams typically run it together with supporting services such as databases, vector stores, automation tools, proxies, and model providers.

That is why OpenClaw is easiest to understand as part of a stack, not as one isolated container.

experiments with local or hosted model providers

OpenClaw is a good example of why self-hosted AI is rarely just an installation problem. To be useful, it usually needs connected services, clearer operations, and an environment the team can evolve over time.

If you want a hands-on path, continue to the OpenClaw practical setup guide.

What OpenClaw is, how teams use it, and why it is best understood as part of a broader self-hosted AI stack.

A practical explanation of OpenClaw and why it is best understood as part of a larger self-hosted AI stack.

What Is OpenClaw?

A deployment is not really done when the command exits successfully. It is done when the system is healthy enough to trust.

That means teams should verify more than “containers started” before they call the work finished.

A better habit is simple: verify health, logs, expected URLs, dependencies, and one real workflow before declaring victory.

If you do not have Stacker installed yet, the quickest path is the install script.

curl -fsSL https://raw.githubusercontent.com/trydirect/stacker/main/install.sh | bash

If you prefer to explore Stacker with Docker first, you can also pull the image:

If your workflow also uses Status Panel locally or on a target server, start here too.

curl -sSfL https://raw.githubusercontent.com/trydirect/status/master/install.sh | sh

GitHub repository and installation steps for Status Panel

do the logs show startup success instead of hidden errors?

do linked services still talk to each other?

does one real user-facing workflow actually succeed?

A lot of painful incidents begin because teams stop at “deployment succeeded” without confirming whether the system is actually usable. Status Panel-style runtime visibility makes this much easier.

If you need to understand the signals behind verification, start with health checks and log-reading. If you need runtime context, continue to the Status Panel explainer.

GitHub repository

How to verify a deployment before calling it done by checking health, logs, dependencies, and one real workflow instead of trusting a success message alone.

A practical guide to verifying health, logs, dependencies, and real workflow success before calling a deployment finished.

How to Verify a Deployment Before Calling It Done

Ansible is a configuration and automation tool that helps teams apply known setup and operational steps to servers in a more repeatable way.

If Terraform is often about creating the infrastructure, Ansible is often about shaping what happens on the machine after it exists.

Ansible matters because many real deployment problems are not only about creating a server — they are about configuring it consistently afterward.

to install and configure software on servers

to reduce configuration drift from manual server edits

to turn infrastructure knowledge into reusable automation steps

How it fits with Terraform and cloud-init

A common chain is: Terraform provisions the infrastructure, cloud-init handles early bootstrap, and Ansible performs richer configuration and deployment tasks.

What Ansible is and why it matters for repeatable server configuration, deployment, and operational setup.

A practical explanation of Ansible and why it matters for repeatable server configuration and deployment workflows.

What Is Ansible?

Terraform is an infrastructure tool that helps teams define and create cloud resources in a repeatable way.

Instead of clicking through cloud dashboards by hand, teams can describe the infrastructure they want and let Terraform create or update it more predictably.

Terraform is often about creating the machine and surrounding cloud resources, not about being the final application runtime itself.

to create servers and cloud resources more consistently

to make infrastructure easier to reproduce later

to keep cloud provisioning closer to a known definition

A common pattern is: Terraform creates the infrastructure, cloud-init can help bootstrap the server, and Ansible handles richer configuration or deployment steps afterward.

What Terraform is and why it matters for creating repeatable cloud infrastructure instead of relying on manual provisioning.

A practical explanation of Terraform and why it matters for repeatable cloud infrastructure in self-hosted deployment workflows.

What Is Terraform?

cloud-init is a way to bootstrap a server with initial setup steps when it first starts.

In practical terms, it helps apply early machine configuration automatically so a new server can become usable faster and more consistently.

cloud-init matters because the first moments of a server’s life often set the tone for everything that happens next.

making fresh servers less dependent on manual first-login work

How it relates to the bigger deployment flow

cloud-init is usually an early bootstrap layer, while tools such as Terraform and Ansible handle broader infrastructure and configuration orchestration around it.

What cloud-init is and why it matters for bootstrapping new servers with more consistent early setup.

A practical explanation of cloud-init and why it matters for bootstrapping new servers more consistently.

What Is cloud-init?

A deployment agent is a runtime component that lives on or near the target server and carries out operational work on behalf of a broader control plane.

In practice, that means the platform can ask the agent to inspect health, fetch logs, restart something, or apply configuration without requiring every action to be performed manually over raw shell access.

A deployment agent helps bridge the gap between “the platform knows about the stack” and “the server can actually do something useful right now.”

they let runtime actions happen closer to the target system

they reduce the need for routine SSH-first operations

they help tie operational actions to a clearer control path

they make health, logs, and other checks easier to standardize

The clearest example here is the Status Panel Agent working alongside Stacker: the control plane can dispatch intent, and the agent can execute the runtime action on the server.

What a deployment agent is and why it matters for runtime actions, health checks, logs, and safer operations after deployment.

A practical explanation of deployment agents and why they matter for health checks, logs, runtime actions, and safer operations after deployment.

What Is a Deployment Agent?

A health check is a way to verify whether a service is actually working well enough to be trusted, not just whether a container happens to be running.

That distinction matters because a service can be technically “up” while still being broken for users if it cannot reach a dependency, answer requests correctly, or complete its real job.

Health checks matter because they turn vague guesses into a more repeatable signal about whether the stack is functioning properly.

that the app responds on an expected endpoint

that a container is not stuck in a restart loop

that dependencies are reachable enough for the service to work

that a key user-facing function still behaves correctly

Health checks are part of how teams move from “deployment succeeded” to “this system is actually healthy enough to operate.” They show up in Status Panel workflows and in practical debugging guides for live stacks.

What a health check is, why it matters in self-hosted stacks, and how it helps teams tell the difference between running and actually healthy.

A practical explanation of health checks and why they matter more than a simple “container is running” signal.