Budibase/budibase · warning

We didn't understand your prompt. This can happen if the pro

Error message

We didn't understand your prompt. This can happen if the prompt isn't specific, or if it's a request for something other than code. Try expressing your request in a different way.

What it means

AIGen.svelte sends the user's prompt to the AI code-generation API (API.generateJs). If the service responds successfully but returns an empty code string — meaning the model couldn't produce code from the prompt — the component throws this descriptive error telling the user to rephrase. The catch block then surfaces it as a notification.

Source

Thrown at packages/builder/src/components/common/CodeEditor/AIGen.svelte:45

  const thresholdExpansionWidth = 350

  $: shouldAlwaysBeExpanded =
    expandedOnly ||
    (parentWidth !== null && parentWidth > thresholdExpansionWidth)

  $: expanded = shouldAlwaysBeExpanded || expanded

  async function generateJs(prompt: string) {
    promptText = ""
    if (!prompt.trim()) return

    previousContents = value
    promptText = prompt
    try {
      const resp = await API.generateJs({ prompt, bindings })
      const code = resp.code
      if (code === "") {
        throw new Error(
          "We didn't understand your prompt. This can happen if the prompt isn't specific, or if it's a request for something other than code. Try expressing your request in a different way."
        )
      }
      suggestedCode = code
      dispatch("update", { code })
    } catch (e: any) {
      console.error(e)

      if ("code" in e && e.code === APIWarningCode.USAGE_LIMIT_EXCEEDED) {
        notifications.error(
          "Monthly usage limit reached. We're exploring options to expand this soon. Questions? Contact support@budibase.com"
        )
      } else if ("message" in e) {
        notifications.error(`Unable to generate code: ${e.message}`)
      } else {
        notifications.error(`Unable to generate code.`)
      }
    }

View on GitHub (pinned to a81a902e9a)

Solutions

1. Rewrite the prompt to explicitly describe the code you want, including inputs and expected output
2. Add context — mention the variable/bindings to use and the framework (plain JS in a Budibase binding)
3. Break the request into a smaller, single-purpose code generation task
4. Retry once in case the empty response was transient; if persistent, the prompt content is the issue

Example fix

// before (prompt)
"make it work"
// after (prompt)
"Write a JavaScript function that takes the binding data.items and returns them sorted by the 'name' property alphabetically"

if (!prompt || prompt.trim().length < 10) {
  throw new Error("Prompt too vague for AI code generation")
}

null

try {
  const resp = await API.generateJs({ prompt, bindings })
  if (!resp.code) throw new Error("AI returned no code — rephrase the prompt")
} catch (e) {
  notify.error(e.message)
}

When it happens

Trigger: Submitting an AI prompt in the builder code editor where the backend returns code === "": vague prompts, prompts not related to code generation, prompts the safety/model layer refuses, or bindings/prompt combinations the model can't satisfy.

Common situations: Asking conversational questions ("what does this do?") instead of requesting code; overly short or ambiguous prompts like "sort it"; prompts in non-English phrasing the model misreads; transient model degradation returning empty output.

Related errors

• AI message content must be a string
• Source table '${relationship.sourceTable}' not found in data
• Target table '${relationship.targetTable}' not found in data
• Invalid custom REST template
• automationId and appId are required

AI-assisted analysis of Budibase/budibase@a81a902e9a (2026-08-29). Data as JSON: /api/errors/c4fb5b83ae07df1b. Report an issue: GitHub.