alibaba/page-agent · warning · InvokeError

CONTEXT_LENGTH

CONTEXT_LENGTH

Error message

Response truncated: max tokens reached

What it means

The model stopped because it hit the max_tokens / output token limit (finish_reason === 'length') before completing a valid response. The full raw response is attached in details, and unlike other errors there is no underlying cause object. For tool-calling agents this usually means truncated, unparseable tool arguments.

Source

Thrown at packages/llms/src/OpenAIClient.ts:154

				InvokeErrorTypes.INVALID_RESPONSE,
				'Response body is not valid JSON',
				error
			)
		}

		const choice = data.choices?.[0]
		if (!choice) {
			throw new InvokeError(InvokeErrorTypes.INVALID_SCHEMA, 'No choices in response', data)
		}

		// Check finish_reason
		switch (choice.finish_reason) {
			case 'tool_calls':
			case 'function_call': // gemini
			case 'stop': // some models use this even with tool calls
				break
			case 'length':
				throw new InvokeError(
					InvokeErrorTypes.CONTEXT_LENGTH,
					'Response truncated: max tokens reached',
					undefined,
					data
				)
			case 'content_filter':
				throw new InvokeError(
					InvokeErrorTypes.CONTENT_FILTER,
					'Content filtered by safety system',
					undefined,
					data
				)
			default:
				throw new InvokeError(
					InvokeErrorTypes.INVALID_SCHEMA,
					`Unexpected finish_reason: ${choice.finish_reason}`,
					undefined,
					data

View on GitHub (pinned to d02db1ee7c)

Solutions

1. Raise or remove max_tokens in the request/config
2. If the provider has a separate output-token cap, request a higher one
3. Reduce expected output size: fewer/simpler tools, shorter requested formats
4. If finish_reason 'length' recurs, catch CONTEXT_LENGTH and re-invoke asking for a continuation or a more compact response

Example fix

// before
const llm = new LLM({ client, maxTokens: 256 })

// after
const llm = new LLM({ client, maxTokens: 4096 })

// keep an eye on prompt+output budget before invoking
const approxInputTokens = estimateTokens(JSON.stringify(request.messages))
if (approxInputTokens + minOutputBudget > modelContextLimit) throw new Error('Prompt too large for safe completion')

const isContextLengthError = (e: unknown): e is InvokeError =>
  e instanceof InvokeError && e.code === 'CONTEXT_LENGTH'

try {
  return await client.invoke(req)
} catch (e) {
  if (isContextLengthError(e)) {
    return await client.invoke({ ...req, max_tokens: (req.max_tokens ?? 1024) * 4 })
  }
  throw e
}

When it happens

Trigger: invoke() where max_tokens (or the provider's default output cap) is reached: very long tool call JSON cut off mid-argument; max_tokens set low in config; reasoning models spending the budget on hidden reasoning tokens before emitting content.

Common situations: Default or user-set max_tokens too small for the task; long structured outputs (big JSON tool inputs); models like o1/gemini-thinking burning tokens on reasoning; combined prompt+completion caps on some providers.

AI-assisted analysis of alibaba/page-agent@d02db1ee7c (2026-08-28). Data as JSON: /api/errors/76836ecc74d5f208. Report an issue: GitHub.