ARIA

Automation Runtime for Intelligent Actions.

Build AI-powered workflows with ARIA.

Your workflows are source-controlled, CI-friendly, and live in YAML files. Define them, run them locally, and inspect every result from saved run artifacts.

Why ARIA?

• Source-controlled workflows: prompts, steps, inputs, and providers live in YAML.
• Full step transparency: see every intermediate result, not just the final answer.
• Extensible runtime: add your own extension when a built-in step is not enough.
• Provider flexibility: OpenAI and local Ollama today, with more providers to come.
• Small runtime: no hosted service, web UI, marketplace, or agent framework.

Install

With the install script:

curl -fsSL https://raw.githubusercontent.com/thrapai/aria/master/install.sh | bash

With optional docling support:

curl -fsSL https://raw.githubusercontent.com/thrapai/aria/master/install.sh | ARIA_PACKAGE='thrapai-aria[docling]' bash

With pipx:

pipx install thrapai-aria

For docling support:

pipx install "thrapai-aria[docling]"

Quickstart

Create a workflow:

aria init

Run it:

ollama pull gemma3:4b
aria run workflow.yml --input path=notes.txt

ARIA prints final outputs as JSON and writes run artifacts under .aria/runs/<timestamp>/.

Check the installed version:

aria version

See upgrade commands:

aria update --method pipx

Workflow

version: "1"
name: summarize_file

providers:
  ollama:
    type: ollama
    base_url: http://localhost:11434

inputs:
  path:
    type: file
    required: true

steps:
  - id: read
    uses: file.read
    with:
      path: "{{ inputs.path }}"

  - id: summarize
    uses: ai.generate
    with:
      model: ollama:gemma3:4b
      prompt: |
        Summarize this file in five concise bullet points:

        {{ steps.read.output.content }}

  - id: save
    uses: file.write
    with:
      path: summary.txt
      content: "{{ steps.summarize.output.text }}"

outputs:
  summary: "{{ steps.summarize.output.text }}"
  summary_file: "{{ steps.save.output.path }}"

Templates can read workflow inputs and previous step outputs:

{{ inputs.path }}
{{ steps.read.output.content }}

Built-In Extensions

• ai.generate
• file.read
• file.write
• utils.json.parse
• utils.text.chunk
• docling.convert - optional

Extension examples live in docs/extensions/.

Providers

ARIA currently supports openai and ollama provider types. A provider entry is an alias that can be used in ai.generate model strings as provider:model.

OpenAI

export OPENAI_API_KEY="..."

providers:
  openai:
    type: openai
    api_key_env: OPENAI_API_KEY
    # Optional:
    # api_key_file: .openai-key
    # base_url: https://api.openai.com/v1

If no providers.openai entry is configured, openai:<model> still works and reads OPENAI_API_KEY from the environment.

Ollama

ollama pull gemma3:4b

providers:
  ollama:
    type: ollama
    base_url: http://localhost:11434

If no providers.ollama entry is configured, ollama:<model> still works and uses http://localhost:11434.

Provider names may be custom aliases when the entry declares a supported type:

providers:
  local:
    type: ollama
    base_url: http://localhost:11434

steps:
  - id: summarize
    uses: ai.generate
    with:
      model: local:gemma3:4b
      prompt: "Summarize: {{ inputs.text }}"

Use providers as provider:model, for example openai:gpt-4.1-mini, ollama:gemma3:4b, or local:gemma3:4b.

Development

uv sync --extra dev
uv run --extra dev python -m pytest
uv run --extra dev ruff check
uv run --extra dev ruff format --check

See CONTRIBUTING.md for contribution and release notes.

License

MIT