Building and Deploying

During local development, your project interacts with the Agistry runtime via the SDK. All adapter logic runs remotely, but you keep full control over flow inputs and execution behavior.

Flow Setup & Adapter Selection

Each flow is made up of adapters that you configure inside Agistry Hub. You can create new adapters through the UI, define their inputs and logic, and organize them under specific projects or tags.

For example, you might define a flow like:

• pdf-text-extractor

• persona-legal

• llm-analyzer

• risk-filter

• telegram-alert

Once configured, you can call them from your app like this:

await agistry.run(
  "Check this contract for risky terms and notify legal ops.",
  ["pdf-text-extractor", "llm-analyzer", "telegram-alert"],
  { docUrl: "https://..." }
);

Flows can be updated in the Hub without needing to redeploy your app — just keep the adapter names stable.

Testing & Debugging

Agistry provides full visibility into flow executions. In the Execution Logs tab of Agistry Hub, you can inspect:

• Input payloads

• Output of each adapter

• Final context state

• LLM prompt & response (if used)

This allows you to debug in real time and refine your adapter configurations iteratively.

We recommend creating test flows (e.g. pdf-text-dev, slack-test) during development and pointing your dev environment at those.

Deploying to Production

When deploying to production:

1. Use a production API key (generated in Agistry Hub)

2. Use production-ready adapters (tagged or versioned accordingly)

3. Avoid exposing the key in frontend code — always call Agistry from your server