microsoft/autogen · error · ValueError

No more mock responses available

Error message

No more mock responses available

What it means

Thrown by ReplayChatCompletionClient.create when the replay buffer is exhausted: _current_index has advanced past the last entry in chat_completions. The replay client is a test double that returns pre-recorded responses in order; once consumed, there is nothing left to return and it raises rather than fabricating output.

Source

Thrown at python/packages/autogen-ext/src/autogen_ext/models/replay/_replay_chat_completion_client.py:176

        return self._create_calls

    async def create(
        self,
        messages: Sequence[LLMMessage],
        *,
        tools: Sequence[Tool | ToolSchema] = [],
        tool_choice: Tool | Literal["auto", "required", "none"] = "auto",
        json_output: Optional[bool | type[BaseModel]] = None,
        extra_create_args: Mapping[str, Any] = {},
        cancellation_token: Optional[CancellationToken] = None,
    ) -> CreateResult:
        """Return the next completion from the list."""
        # Warn if tool_choice is specified since it's ignored in replay mode
        if tool_choice != "auto":
            logger.warning("tool_choice parameter specified but is ignored in replay mode")

        if self._current_index >= len(self.chat_completions):
            raise ValueError("No more mock responses available")

        response = self.chat_completions[self._current_index]
        _, prompt_token_count = self._tokenize(messages)
        if isinstance(response, str):
            _, output_token_count = self._tokenize(response)
            self._cur_usage = RequestUsage(prompt_tokens=prompt_token_count, completion_tokens=output_token_count)
            response = CreateResult(
                finish_reason="stop", content=response, usage=self._cur_usage, cached=self._cached_bool_value
            )
        else:
            self._cur_usage = RequestUsage(
                prompt_tokens=prompt_token_count, completion_tokens=response.usage.completion_tokens
            )

        self._update_total_usage()
        self._current_index += 1
        self._create_calls.append(
            {

View on GitHub (pinned to 027ecf0a37)

Solutions

1. Add more recorded responses to chat_completions covering every expected turn, including tool-call rounds
2. Reduce the loop: set max_turns on the agent so it stops within the scripted budget
3. Create a fresh ReplayChatCompletionClient per test (or reset _current_index) so state does not leak between cases

Example fix

# before
replay = ReplayChatCompletionClient(["Hello", "How can I help?"])
# agent loop runs 3 LLM calls -> third raises

# after
replay = ReplayChatCompletionClient(["Hello", "How can I help?", "Anything else?"])
# or: await agent.run(task, max_turns=2)

class CountingReplayClient(ReplayChatCompletionClient):
    def remaining(self) -> int:
        return len(self.chat_completions) - self._current_index

# before each agent run:
assert replay.remaining() >= expected_llm_calls, f"scripted responses exhausted: {replay.remaining()}"

def replay_has_responses(replay: ReplayChatCompletionClient, needed: int = 1) -> bool:
    return replay._current_index + needed <= len(replay.chat_completions)

try:
    result = await replay.create(messages)
except ValueError as e:
    if "No more mock responses" in str(e):
        pytest.fail("replay buffer too small: add scripted completions or lower max_turns")
    raise

When it happens

Trigger: Calling create (or letting an agent loop call it) more times than there are recorded completions passed to the constructor. Each create call consumes exactly one entry and increments _current_index.

Common situations: Agent loops in tests that iterate more turns than scripted (a tool-call round-trip adds an extra completion); sharing one replay client across multiple test cases without resetting; forgetting that every create counts, including intermediate tool-call responses.

Related errors

• Failed to load recorded session: '{self.session_file_path}':
• No more recorded turns to check.
• Recorded call type mismatch at index {self._record_index}: e
• Current message list doesn't match the recorded message list
• Early termination. Only {self._num_checked_records} of the

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/2185858268e35d5a. Report an issue: GitHub.