BANDURA Download

Local AI workflow builder

Connect your model
to the rest of your work.

Use Bandura to build workflows with AI models, API calls, and scripts. Edit the steps visually, inspect their results, and save the whole workflow as a readable YAML file.

Download Bandura free Start with a local example

No account required · macOS, Windows and Linux · Stable v1.0.0

The source of a support-classification workflow, three expected categories, the last sample’s output, and three passed case results The source of a support-classification workflow, three expected categories, the last sample’s output, and three passed case results
Demonstration results: three synthetic messages checked against their expected categories using a deterministic local server. This shows the evaluation workflow, not model accuracy. Development preview of the redesigned app.

Where it fits

A few useful starting points.

Choose a small task with an output you can check. Add more steps once that part works.

Classify incoming messages

Ask your model to assign a category, check it against an allowed list, and route the result. Test labelled samples before working with real messages.

Download the classification example

Prepare a draft with context

Fetch information from an API, include it in a model prompt, and validate the output before you review it. Keep customer replies behind a review step.

Configure an AI step

Check an API sequence

Sign in, create a record, then fetch it back. Pass values between requests and use assertions to catch unexpected responses.

Build a request and a check

Try it without an AI key

Run a local example first.

The starter guide includes an order calculation with an assertion. It runs on your computer without network access or a model account.

Open the starter guide

  1. Create a workflow

    Save the guide’s example as order-process.aether in your workspace.

  2. Run and inspect

    The script calculates 32. The assertion checks the total. Select a step to inspect the result and variables.

  3. Try a failing check

    Change the expected total to 99 and run again. Inspect the failed assertion, then restore the expected value.

When you add AI

Choose your model. Check its output.

Open Settings → AI, select a provider or compatible endpoint, set a model, and test the connection. Provider usage charges apply to your own account.

Before you start.

What is an AI workflow?

An AI workflow connects an input to a sequence of steps: model calls, API requests, scripts, and checks. Instead of copying results between tools by hand, you save the sequence and run it with new inputs. Model output can vary between runs, so checks and human review still matter.

Do I need to write code?

You can add and connect steps on the visual canvas. Configure prompts, requests, and variables in the inspector. Assertions use JavaScript expressions, and script steps use JavaScript, so some workflows need code. The YAML view lets you edit the same file directly.

Can I use Ollama or LM Studio?

Yes. Configure an OpenAI-compatible local endpoint and a model in Settings → AI. Anthropic and Gemini also have native adapters. Hosted providers receive your prompts and supplied context; a local model keeps those model calls on your machine. Requests to remote APIs and tools still cross the network.

Can I run workflows on a schedule?

The desktop app runs locally when you start a workflow. It does not include scheduling or hosted execution. You can use your own scheduler or CI runner with the separately versioned CLI; check its documentation for supported nodes.

Is Bandura free for work?

Yes. Bandura is free for personal and commercial use, with no paid seats or feature tiers and no account required. You supply your own model and service accounts, and pay any charges from those providers. The app is proprietary software; your workflow files remain yours.

Try Bandura with a sample workflow.

Free for personal and commercial use. No signup or payment.

Download Bandura