Describe the architecture once.Get the infrastructure code for each platform you run.
Aggrega is an open-source compiler for infrastructure as code: one YAML description of your application becomes Kubernetes, Helm, Compose, Terraform or CloudFormation files you review and own.
Already built an agent with LangGraph, CrewAI, n8n, Dify and more? Import it.
Linux and macOS
curl -fsSL https://aggrega.ai/install.sh | shWindows, in PowerShell
irm https://aggrega.ai/install.ps1 | iexTwo quickstart examples, built up one step at a time: an AI agent on a local model, and a web app with PostgreSQL. Each step shows the files it changes with the lines it adds, the aggrega build command, and the files the build writes.
aggregate.yaml
kind: Aggregatespec: components: chat: type: core/compute/server agent: type: core/ai-agent/langgraph llm: type: core/ai-model/ollama memory: type: core/datastore/valkeyaggrega build --release release.yaml --auto-pull
- dist/compose/docker-compose.yaml
Works with your AI assistant
Paste this prompt into an assistant that can run commands, and it installs Aggrega, adds the Aggrega skill to itself and connects to the Aggrega MCP server. The prompt has it check with you before it installs anything.
Set up Aggrega for me, one step at a time, and tell me what you ran at each step.
1. Run `aggrega version`. If the command is not found, ask me before you install Aggrega, then install it for my user with the official installer: on Linux or macOS `curl -fsSL https://aggrega.ai/install.sh | sh`, on Windows in PowerShell `irm https://aggrega.ai/install.ps1 | iex`. Run `aggrega version` again. If the installer printed a line to add to my `PATH`, show it to me.
2. If `https://aggrega.ai/docs/ai-assistants/skill/` gives a way to install the Aggrega plugin in you, ask me before you install it, then install it that way, or tell me what to do where it is a step I take in your interface. The plugin adds the Aggrega skill and connects you to the Aggrega MCP server. Otherwise, run `aggrega skills install --help` and find your own identifier in the list it prints. Ask me whether I want the skill in this project or in every project, then run `aggrega skills install --agent <your identifier>`, adding `--global` for every project. If you are not in the list, skip installing the skill and say so.
3. If you installed the plugin, you are connected to the Aggrega MCP server already. Otherwise, connect yourself to the Aggrega MCP server, which is the command `aggrega mcp`, the way you add an MCP server, as `https://aggrega.ai/docs/ai-assistants/mcp-server/` describes for you. If you need a restart or a reload to load the plugin or the server, tell me what to do.
4. Once you are connected, call the `describe-capabilities` tool with no arguments and summarize what it lists.
- The skill
- Tells the assistant how the authoring loop runs, and to take every answer about a schema or a file from the aggrega binary.
- The MCP server
- aggrega mcp gives the assistant Aggrega's tools: scaffold, validate, build and the rest.
How it works
You describe the application, Aggrega writes the files, and the tools you already use deploy them. Aggrega never reaches a running cluster or a cloud account.
- Kubernetes
- Helm
- Compose
- Terraform
- CloudFormation
files you review and own
Describe the application
An Aggregate names its components and how they connect.
Choose the environment
A Release fills in that environment's settings.
Pick how each part is built
A Manifest maps each component to a Recipe for your platform, such as a Helm chart, a Terraform module or a container.
Run aggrega build
It writes the files for each platform: the same files for the same input, every time.
Deploy with your tools
What it wrote is deployed by ArgoCD, Flux or Terraform.
What you can build
Start from a runnable example, or from the catalog's components.
AI agents and gateways
Agents built with LangGraph, CrewAI, LlamaIndex, Dify, n8n and more, behind model gateways with token budgets, semantic caching, log masking and a prompt-injection shield.
An agent behind a gateway with a token budgetTraditional workloads
Databases, caches, message brokers and identity, for example PostgreSQL, MariaDB, Redis, Kafka and Keycloak.
One PostgreSQL database, deployed three waysFeatures
Capabilities you attach per environment without touching the application, for example a prompt-injection shield, masking of sensitive data, HTTPS ingress, monitoring and backups.
A gateway that screens each request for prompt injectionOne application, many platforms
The same Aggregate builds for Docker Compose on a laptop and Kubernetes in production: changing target changes a mapping, not your code.
One Aggregate, built for Compose and Kubernetesaggregate.yamlwritten onceOn a laptop
release-compose.yamlmanifest-compose.yamlaggrega builddocker-compose.yamlOn a Kubernetes cluster
release-helm-synth.yamlmanifest-helm-synth.yamlaggrega buildKubernetes manifestsEach platform is built from its own Release and Manifest. The Aggregate, which describes your application, stays the same. Governance and policies
Your policies plug into the compiler and stop a build that breaks them, saying what to fix.
One policy, checked on four targetsaggrega buildyour policyplugged into the compilerpassfiles writtenready to review and deployfailbuild stoppedpolicy plugin "rego" rejected the build: terraform/modules/db/main.tf: the RDS instance db stores its data unencryptedA policy can stop a build and say what to fix. It never changes your files to make them pass. Reproducible builds
Same input, same output, with every dependency pinned in a lockfile.
Build an example and its lockfileThe standard-library catalog
Components for agents, models, gateways, datastores, messaging, observability, security and storage, ready to use in an Aggregate.
Browse the catalog
Your assistant, with Aggrega
What an assistant that writes your infrastructure code gains by working through Aggrega.
- Reproducible
- Same input, same files, with every dependency pinned in a lockfile.
- One description
- One description of the application, built for every platform it runs on.
- Policies at build time
- Your policies run in the compiler and stop a build that breaks them.
- A signed, versioned catalog
- What an assistant pulls from the catalog is versioned and signed, and the pull verifies each signature.
- Checked by the compiler
- The assistant validates and builds with the aggrega binary, and takes its answers about schemas and files from it.
Targets
Each platform gets the files it reads natively. AI agents run on any Kubernetes cluster, and many on Compose too.
- Kubernetes
- Kubernetes manifests, one file per resource, for ArgoCD or Flux to apply.
- Helm
- Helm charts, rendered into Kubernetes manifests.
- Compose
- A Docker Compose file, to run the application on a laptop.
- Terraform
- Configuration for managed cloud services. Aggrega never runs Terraform.
- CloudFormation
- Stack templates for AWS. Aggrega never runs CloudFormation.
Import
aggrega import reads a project in one of these formats and writes draft Aggrega files for you to review and edit.
- LangGraph
- CrewAI
- n8n
- Dify
- Langflow
- Flowise
- Google ADK
- Microsoft Agent Framework
- MCP client configurations
- Compose
Get started
Quickstart
Install Aggrega, build a runnable example, run it on your machine, then grow it one step at a time.
Getting startedRun a first example
Each runnable example is a complete workspace you build with the aggrega CLI.
Browse the examplesInstall
One command installs the aggrega binary for your user, with no administrator rights.
Install the binary
Open source
The Aggrega compiler is open source under Apache-2.0, and stays that way: there will be no closed edition of the compiler, no open-core. All the features of the compiler will be available in the open source version.
The compiler
The aggrega binary and its command-line interface.
aggrega-ai/aggrega-cli on GitHubVisual Studio Code
Validation, navigation, hover and completion for Aggrega YAML, backed by the aggrega binary.
The extension on GitHub