Budibase/budibase · warning · Error

Invalid tool call summary response

Error message

Invalid tool call summary response

What it means

generateToolCallSummary asks the LLM for a short plain-text summary of a single tool call and normalizes it. If the normalized summary is empty (empty text, or only quotes/whitespace), it throws Error("Invalid tool call summary response"). Called from recordToolCall when recording each tool call on an agent request timeline.

Source

Thrown at packages/server/src/sdk/workspace/ai/agentRequests/helpers.ts:237

  const result = await generateText({
    model: llm.chat,
    providerOptions: llm.providerOptions?.(false),
    headers: {
      "x-litellm-tags": "bb-agent-request-tool-call-summary",
    },
    instructions:
      'Summarize a single tool call for a UI timeline entry, in plain user-friendly language, not technical jargon - describe what the action did and its outcome, e.g. "Searched tickets for email server errors" or "Failed to update the customer\'s address". Return plain text only. Use at most 6 to 7 words, no quotes, no punctuation unless necessary.',
    prompt: JSON.stringify({
      tool: readableName || toolName,
      status,
      input,
      output,
    }),
  })

  const summary = normalizeTitle(result.text || "")
  if (!summary) {
    throw new Error("Invalid tool call summary response")
  }

  return summary
}

export async function generateInteractionSummary({
  latestPrompt,
  agentId,
  sessionId,
}: {
  latestPrompt: string
  agentId: string
  sessionId: string
}): Promise<string> {
  const agent = await sdk.ai.agents.getOrThrow(agentId)
  const llm = await sdk.ai.llm.createLLM(
    agent.aiconfig,
    sessionId,

View on GitHub (pinned to a81a902e9a)

Solutions

1. Catch the error and fall back to a static summary built from the tool name and status (e.g. "Ran readableName — success")
2. Check LiteLLM/provider logs for empty or filtered completions and fix the aiconfig
3. Truncate very large tool input/output before prompting to avoid degenerate completions

Example fix

// before
const summary = await generateToolCallSummary({ toolName, status, ... })
// after
let summary
try {
  summary = await generateToolCallSummary({ toolName, status, ... })
} catch {
  summary = `${readableName || toolName} ${status === "success" ? "completed" : "failed"}`
}

let summary
try {
  summary = await generateToolCallSummary({ toolName, status, input, output, agentId, sessionId })
} catch (err) {
  if (err instanceof Error && err.message === "Invalid tool call summary response") {
    summary = `${readableName || toolName} ${status}`
  } else {
    throw err
  }
}

When it happens

Trigger: The LLM returns an empty, filtered, or whitespace-only completion when summarizing a tool call (e.g. very large/empty tool output, provider outage yielding empty body).

Common situations: Content filters suppressing output for sensitive tool input/output; provider misconfiguration in the agent's aiconfig; oversized prompts leading to empty completions.

Related errors

• Invalid agent request title response
• No response found
• Invalid agent request link response
• Invalid agent request link response: unknown requestId "${de
• Unable to retrieve object

AI-assisted analysis of Budibase/budibase@a81a902e9a (2026-08-29). Data as JSON: /api/errors/3af61ece05324333. Report an issue: GitHub.