Skip to main content
Use Aster in FlowRunner to generate text or run a Chat Completions request inside a workflow. The integration offers two actions: Generate Text and Create Chat Completion. Both return a complete response without streaming. View the Aster integration in FlowRunner.

​
Before you start

You need a FlowRunner account and an AsterWise account with prepaid balance and an active API key. Follow the AsterWise Quickstart to set up your account and key. Keep the key private; do not put it in prompts, screenshots, or shared workflow examples. Choose the model for your task:
  • Aster Work (aster-work) for writing, research, and everyday tasks
  • Aster Code (aster-code) for coding tasks and reasoning about code you provide
Start with synthetic data so you can check the setup before using real customer or business information.

​
1. Add Aster to your flow

In the FlowRunner editor, open Extensions → Aster and add Generate Text to your flow. Configure the Aster extension with your AsterWise API key. Aster is an action in the workflow. To automate it later, connect it to a trigger from another service or a schedule.

​
2. Configure a small text request

Use these settings for an initial example:
  • Model: Aster Work
  • System Prompt: Draft concise customer replies using only the supplied policy. Do not invent facts or make commitments.
  • Prompt: Customer question: What time do you open on Saturday? Policy: Our Saturday opening hours are 10 am to 4 pm. Draft a friendly reply in one sentence.
  • Reasoning Effort: low
  • Max Completion Tokens: 2048
These are example settings, not a guaranteed output allowance for every task. Reasoning and visible answer text share the token cap. Increasing the cap allows more billable output on a successful request. Choose run block to send the request. A successful test request uses your prepaid balance.

​
3. Review the response

Generate Text returns the generated text, finish reason, token usage, reported model, and completion ID. Review the result before connecting it to another action. For this example, check that the reply accurately gives the Saturday opening hours and adds no unsupported information. Wording can vary. Check the finish reason too:
  • stop indicates normal completion; still review the content for accuracy
  • length means the output limit was reached; treat the answer as potentially incomplete
A completed response with length and recorded usage is billable, even if the visible answer is empty. The usage values are token counts, not the dollar charge.

​
Add a human-reviewed workflow

Once your first request works, you can build on it. For example, a customer-reply workflow can retrieve an email, draft a reply with Aster Work, and save the reply as a Gmail draft. A person should check the recipient, facts, tone, and any commitments before sending it. For coding work, pass a small bug report and the relevant code to Aster Code. Keep file changes, test execution, and review in your own tools or workflow. Review the proposed change and actual test results before accepting it.

​
Use Create Chat Completion for more control

Choose Create Chat Completion when you need conversation history or function-tool requests. It returns the complete response, first-choice text, tool calls, finish reason, and usage. Aster returns tool requests; your workflow executes them. To continue a tool conversation, include the assistant’s tool-call message, a tool-result message with the matching tool_call_id, and the conversation history in the next request. Review and authorize consequential actions before executing them. Use the Chat Completions reference for the API request and response shape, and Parameters and model compatibility for supported controls. These references describe the API; configure the corresponding inputs through FlowRunner’s action editor.

​
If the request fails

  • Authentication or balance error: Check that the key is active and your AsterWise account has sufficient balance
  • Unsupported model or setting: Use aster-work or aster-code and check the parameter reference before retrying
  • provider_empty_response: The failed request is not charged. A larger output-token cap or lower supported reasoning effort may help, but neither guarantees an answer
  • Timeout or unclear outcome: Check Usage before submitting again
See Troubleshooting. If you need support, include the error and request identifier when available. Never share your API key.