Build the process.
See every step.

Build processes that combine AI, APIs, scripts, conditions, and explicit result checks. Examples include classifying requests, extracting structured data, preparing drafts, and validating an order process. Workflows run when you start them in the desktop app or through the CLI or MCP server.

A workflow makes the inputs, steps, outputs, and checks explicit in a file you can edit and run again. AI can help build the process and execute individual steps; the graph shows how those steps connect.

Technical builders and small teams who want to own their workflow files, combine AI with APIs, and inspect execution locally. Start with one repeatable process and sample inputs.

Built-in scheduling and hosted execution are not available today. The desktop runner is local. The CLI can be invoked by a scheduler or CI system you manage; that environment must supply its own dependencies and credentials.

The development build adds a choice node that sends explicit options to an endpoint, validates the returned label and probabilities, and stores the result for later conditions. A local Laya adapter is an optional example. Jev is not a built-in provider today.

See the workflow guide →

The development desktop build adds an MCP Connections page. Save access to a remote server, check which saved apps expose tools, then add one tool to an open workflow or copy a YAML step. Checking reads the tool list without running a tool. Registry listings do not connect accounts; OAuth login and local stdio servers are not built in yet. The published release already supports direct HTTP APIs.

Connect an MCP app →

One workflow stored as readable YAML with embedded JavaScript. The visual graph and code editor are two views of that file. Commit it to Git or share it with the supporting scripts and datasets it uses.

Define explicit assertions and try labelled sample cases. A completed AI call only means a response was returned; passing your checks does not guarantee factual correctness. Inspect the outputs and improve the checks for your process.

Build and evaluate a workflow →

Yes. Attach inline sample rows or a CSV/JSON dataset. Each row runs independently, and assertions can compare an actual result to that row's expected value. The development build keeps each completed row's verdict and duration in History, where you can search failures and copy results. It does not yet keep a full per-case node trace.

The recipient opens the file or project in Bandura and supplies their own model settings and credentials. Include required scripts, subflows, and datasets. Keep secrets out of the shared files. External API and model availability can affect results.

Yes. Requests, assertions, imports, and the existing CLI remain supported. API testing is one workflow use case; the main product focus is AI workflow automation.

Connect Anthropic, Google Gemini, or an OpenAI-compatible endpoint. Local runtimes such as Ollama and LM Studio can power AI steps, chat, and generation. Use the connection check in Settings before running a workflow.

Editing, local scripts, and stored history work offline. Requests need access to their target services. Remote AI calls send prompts and supplied context to your configured provider; a model running locally can avoid that external call. The optional release check can be disabled in Settings.

You use your own provider account and pay its usage costs directly. Bandura adds no AI markup or credit system. A local model does not incur a hosted-provider charge, but uses your own hardware.

Yes. The current release candidate is available for macOS, Windows and Linux. macOS builds are unsigned early access; the download page explains first-launch approval. Some new workflow improvements described in the development guide are being prepared for the next release.

Yes, for personal and commercial use. No trial, paid seat, account, activation key, or subscription is required.

Not today. The .aether format is documented, and your workflow files remain yours. The installed application and runner do not require an activation server.

Patreon support is optional and unlocks no extra features.

Support on Patreon →