Skip to content

Install ​

On your own computer, Curio runs as three servers: a backend that stores projects, a sandbox that runs node code, and the web app you open in your browser. You can install it with pip or run it in Docker, on Windows, macOS or Linux. A local Curio signs you in automatically, as a single shared guest user, and opens on the projects page.

Install with pip ​

Create a fresh Python environment (with conda, for example) with the Python version Curio supports; the installation guide linked at the end of this page names it. Then install and start Curio:

bash
pip install utk-curio
curio start

curio start first installs the Python libraries the built-in nodes use, so the first start takes a few minutes. When the servers are up, open http://localhost:8080; press Ctrl+C in the terminal to stop them. Start Curio from the same folder each time: it keeps its working files in a .curio folder there, and relative file paths in node code start from there too.

Nodes that run JavaScript, such as JS Computation, use the Node.js installed on your computer. Curio starts without Node.js, but if the installed one is too old it stops and says which version to install.

Run it with Docker ​

With Docker installed, clone the repository and start the stack:

bash
git clone https://github.com/urban-toolkit/curio.git
cd curio
docker compose up

The first run builds the image, which takes a while; docker compose up --build rebuilds it after you update the clone. Then open http://localhost:8080. The image includes everything Curio needs, Node.js too, and Curio starts with the example dataflows on the projects page.

To change Curio itself, run it from a clone with python curio.py start instead (see Contribute).

Commands and flags ​

curio --help lists every command and flag. The ones you are most likely to use:

  • curio start starts all three servers; curio start backend, sandbox or frontend starts just one.
  • curio setup installs the Python libraries Curio and its node packages need, then exits.
  • --no-project skips the projects page and opens the canvas directly, for demos.
  • --deploy turns on accounts and the sign-in page, for a shared server. It needs a host that can isolate each user's node code, such as the Docker image, and refuses to start on one that cannot (see Run your own server).
  • --frontend-port, --backend-port and --sandbox-port move the servers off ports 8080, 5002 and 2000.

Set up the AI features ​

Curio ships with no AI provider. Its agents, node-authoring assistants and chat answer through an LLM configuration you set up, and Curio sets no spending limit: the provider bills the key's owner.

To add a configuration:

  1. Get an API key from the provider: OpenAI, Anthropic or Gemini. A model you run yourself, such as one in Ollama or LM Studio, may need no key.
  2. Open API Settings from the top bar. On the Projects and catalog pages it opens the settings page; on the canvas and the dashboard it opens as a drawer on the right.
  3. On the API keys tab, click + (Add configuration) at the end of the table's header.
  4. In Kind, choose Language model.
  5. Give it a Label, and pick the Provider: OpenAI, Anthropic, Gemini, or Custom for any OpenAI-compatible endpoint (Custom asks for a Base URL).
  6. Paste the API key, and choose the Model: Fetch models suggests what the endpoint serves, or type a model name.
  7. Click Add configuration. The configuration's row shows in the list, with Key saved under its name when you gave a key.

On a local Curio you are the shared guest, and everyone using that Curio shares the configurations you save. AI agents covers the rest.

You can instead give Curio endpoints of its own when you start it, in a file with one section per endpoint:

ini
[Lab gateway]
base_url=https://llm.example.org/v1
api_key=your-api-key
models=gpt-4o-mini,gpt-4o
bash
chmod 600 llm-endpoints
curio start --llm-endpoint-file llm-endpoints

Each section needs api_key and models. base_url is the endpoint, left out for the provider's own, and provider is openai_compatible, the default, which covers OpenAI and any OpenAI-compatible server such as Ollama or LM Studio, or anthropic or gemini. A keyless server still needs a placeholder key. Every account then sees a Curio row for each endpoint on the API keys tab and can pick its models on the Agent configuration tab, and the first endpoint's first model answers until an account picks another. With Docker, mount the file into the container (see Run your own server).

On the hosted instance, or on a server started with --deploy, each account adds its own configurations in API Settings or picks one of the server's models, and a guest answers on the server's first model.