Aider-AI/aider · error · Exception

No data found in LLM response!

Error message

No data found in LLM response!

What it means

Raised in BaseCoder's response-processing path when BOTH attribute probes on the LLM completion message failed: completion.choices[0].message.tool_calls raised AttributeError AND completion.choices[0].message.content raised AttributeError. That means the chat completion object has neither tool calls nor textual content, so there is nothing to show or apply. It is a plain Exception (not a typed error), signalling an empty/malformed API response.

Source

Thrown at aider/coders/base_coder.py:1880

            except AttributeError:
                reasoning_content = None

        try:
            self.partial_response_content = completion.choices[0].message.content or ""
        except AttributeError as content_err:
            show_content_err = content_err

        resp_hash = dict(
            function_call=str(self.partial_response_function_call),
            content=self.partial_response_content,
        )
        resp_hash = hashlib.sha1(json.dumps(resp_hash, sort_keys=True).encode())
        self.chat_completion_response_hashes.append(resp_hash.hexdigest())

        if show_func_err and show_content_err:
            self.io.tool_error(show_func_err)
            self.io.tool_error(show_content_err)
            raise Exception("No data found in LLM response!")

        show_resp = self.render_incremental_response(True)

        if reasoning_content:
            formatted_reasoning = format_reasoning_content(
                reasoning_content, self.reasoning_tag_name
            )
            show_resp = formatted_reasoning + show_resp

        show_resp = replace_reasoning_tags(show_resp, self.reasoning_tag_name)

        self.io.assistant_output(show_resp, pretty=self.show_pretty())

        if (
            hasattr(completion.choices[0], "finish_reason")
            and completion.choices[0].finish_reason == "length"
        ):
            raise FinishReasonLength()

View on GitHub (pinned to 5dc9490bb3)

Solutions

1. Log the raw completion (self.verbose / print(completion)) to see which fields the provider actually returned.
2. Verify provider compatibility: the endpoint must return OpenAI-shaped messages with content or tool_calls on choices[0].message.
3. Retry the request — transient empty responses from proxies/self-hosted servers often succeed on resend.
4. If using a custom LiteLLM/aider model mapping, check the model config (API base, model name) points at a truly OpenAI-compatible API.
5. Upgrade aider and the openai SDK; schema handling of reasoning_content vs content changed across versions.

Example fix

# before
# rely on the raw exception
try:
    coder.run_one(user_msg)
except Exception as e:
    print(e)  # "No data found in LLM response!"

# after
# inspect what the provider actually returns before blaming the prompt
completion = client.chat.completions.create(...)
msg = completion.choices[0].message
assert getattr(msg, "content", None) or getattr(msg, "tool_calls", None), completion.model_dump()

def completion_has_data(completion) -> bool:
    try:
        msg = completion.choices[0].message
    except (IndexError, AttributeError):
        return False
    return bool(getattr(msg, "content", None)) or bool(getattr(msg, "tool_calls", None))

def is_usable_completion(c) -> bool:
    """True if the completion carries content or tool calls."""
    msgs = getattr(getattr(c, "choices", None) or [None], "__getitem__", lambda i: None)(0)
    m = getattr(msgs, "message", None)
    return bool(m) and (bool(getattr(m, "content", None)) or bool(getattr(m, "tool_calls", None)))

try:
    coder.send(new_message)
except Exception as e:
    if str(e) == "No data found in LLM response!":
        # empty/malformed provider response — safe to retry once, then report
        coder.send(new_message)
    else:
        raise

When it happens

Trigger: A chat completion arrives where choices[0].message lacks both 'content' and 'tool_calls' — typically from a non-OpenAI-compatible gateway, a response truncated by length limits, a content-filtered response, or a caching/mock layer returning a bare object. Also hit when the response represents only reasoning/abort fields with no message payload.

Common situations: Using an OpenAI-compatible proxy or local model server whose schema drifts from the OpenAI SDK (missing attributes); empty responses from content filters; streaming assembled into an incomplete message; bugs in custom LLM middleware or recorded-fixture replays in tests.

Related errors

• Unknown edit format {edit_format}. Valid formats are: {', '.
• # {len(failed)} SEARCH/REPLACE {blocks} failed to match!\n
• Bad/missing filename. The filename must be alone on the line
• Expected `=======`
• Expected `>>>>>>> REPLACE` or `=======`

AI-assisted analysis of Aider-AI/aider@5dc9490bb3 (2026-08-15). Data as JSON: /api/errors/22068bbf27bbfe25. Report an issue: GitHub.