`Invalid response from LLM call - None or empty.` — the CrewAI call that came back with nothing

Received None or empty response from LLM call. ValueError: Invalid response from LLM call - None or empty.

CrewAI asked the model for its next step and got back nothing it could use: no text, and nothing else that counts as an answer. The error is not about your prompt or your task; it is a guard that refuses to carry an empty answer forward. What you want to know is which of the two places in the source raised it, and why the answer was empty. Every claim below is read off CrewAI’s own source at release tag 1.15.26, fetched on 9 October 2026, with the address of each quoted line given; where we draw a conclusion from those lines rather than quote them, we say so.

Where it is raised: lib/crewai/src/crewai/utilities/agent_utils.py, in two functions — the ordinary model call, and the forced final answer after max_iter.

What it means: the value returned by llm.call(...) was None, an empty string, or (in the ordinary call) anything else Python treats as false.

First check: the traceback. If it passes through handle_max_iterations_exceeded, your run had already hit its iteration cap.

Where the string comes from

The ordinary path is get_llm_response (and its async twin, aget_llm_response). Both call the model and hand the result to _validate_and_finalize_llm_response, which opens like this (agent_utils.py, lines 527–533):

if not answer: if verbose: printer.print( content="Received None or empty response from LLM call.", color="red", ) raise ValueError("Invalid response from LLM call - None or empty.")

The second path is the one that runs after an agent reaches max_iter: handle_max_iterations_exceeded makes one more call asking for a final answer, and checks the result with a narrower test (agent_utils.py, lines 403–409):

if answer is None or answer == "": ... raise ValueError("Invalid response from LLM call - None or empty.")

Three things follow from those lines.

It is a plain ValueError. There is no dedicated exception class to catch; the only thing that tells it apart is the message text.
The red line comes first. With verbose on, CrewAI prints Received None or empty response from LLM call. just before raising, so that line in your log marks the exact call that failed.
Your after-call hooks never see it. In _validate_and_finalize_llm_response the emptiness check comes before the after-LLM-call hooks run, so a hook cannot rescue or rewrite an empty answer (our reading of the order of the code above).

What makes the answer empty

According to the docstring on LLM.__new__ in llm.py, CrewAI routes a model from a native provider (OpenAI, Anthropic, Azure, Bedrock, Gemini) to that provider’s own SDK class, and otherwise falls back to LiteLLM. We read only the LiteLLM path this time, so the two roads below are shown for that path; the native classes were not checked.

The model returned no text. In the non-streaming handler, the text is taken from the provider’s message with a fallback to an empty string (llm.py, line 1274):

text_response = response_message.content or ""

When there are no tool calls either, that empty string is what the function returns, and the guard above turns it into this error. A provider message whose content is None therefore arrives here as "", not as None.

The model asked for a tool and the tool raised. When the model returns tool calls and CrewAI has the functions to run them, it runs the first one. If that function raises, the handler logs it and gives up (llm.py, line 1755):

logging.error(f"Error executing function '{function_name}': {e}")

and then returns None. Back in the response handler, a None tool result is skipped and the function falls through to return the message text (llm.py, lines 1309–1326):

if tool_result is not None: return tool_result ... return text_response

If the model sent no text beside its tool call, that text is "", so (our inference) a tool that throws surfaces as this error, with the tool’s own exception only in the log. If you see Error executing function just above the red line, that is your cause.

Streaming is different. On the streaming path of the same class, an empty stream is caught before this guard, with a different message (llm.py, lines 931–933):

raise Exception( "No content received from streaming response. Received empty chunks or failed to extract content." )

That check fires only when the stream brought neither text nor tool calls, so on the LiteLLM path a completely empty stream stops with that message rather than with this page’s string (our reading of the two paths).

The full-text export of CrewAI’s documentation, docs.crewai.com/llms-full.txt, which we fetched on 9 October 2026 at 1,932,254 bytes, contains the phrase None or empty zero times.

What to do

Read the traceback for the function name. Through handle_max_iterations_exceeded: the agent ran out of iterations and even the forced final answer came back empty — fix the cap problem first (the page on max_iter below). Through get_llm_response or aget_llm_response: an ordinary step came back empty; keep going.
Look at the line above the red one. Error executing function ‘…’ means a tool raised. Make that tool catch its own errors and return a string that says what went wrong, so the model gets an answer it can act on instead of nothing.
Look at what the provider sent. With no tool error, the model returned a message with no text. The call events CrewAI emits for that response carry its finish_reason (the handler passes it along in the lines linked above); that field, not the error, tells you whether the provider stopped on length, on a filter, or on something else.
Retry only once you know which. If a tool threw, the same input will most likely make it throw again (our inference), so fix the tool before you rerun the task.

The thing worth keeping even if you never see this string again: an agent framework has to put something in the transcript after every model call, and “nothing” is the one answer it cannot carry. Most of the time the nothing was made earlier — by a tool that threw, or by a run that had already used up its steps — and the error is only where it was noticed.

What is behind this site

There is a written guide: the step-and-turn arithmetic as a formula you can run against a brief before you launch it, why raising a cap does not finish the job, and batch.py, one standard-library file that collapses a per-item loop into a single pass — fewer turns, which means fewer final calls made from an unfinished transcript when a run hits its cap. It is $19, on a storefront that delivers the files automatically and carries a 30-day money-back guarantee (checked 7 October 2026). One working way to pay today is 19 USDC on Base, and delivery is manual: you email the transaction hash and the files come back as a reply. This page is free, ungated, and sells nothing on its own.

The short version: Invalid response from LLM call - None or empty. is a plain ValueError raised in CrewAI’s agent_utils.py when llm.call(...) returns None or an empty string — either on an ordinary step, or on the forced final answer after max_iter. On the LiteLLM path, a message with no text becomes "", and a tool that raises leaves the call returning that empty text. Read the traceback for which function raised it, look for Error executing function just above, and check the response’s finish_reason before you retry.

Nearby

Maximum iterations reached. Requesting final answer. — the CrewAI line printed just before the forced final call, the second of the two places this error is raised.
recursion_limit, max_iter, max_turns: one wall under four names — what each framework calls the step ceiling and what each does on contact.
AI_NoOutputGeneratedError — the AI SDK’s run that ended on a tool-call step and left no output, an empty result seen from another library.
max_turns vs timeout vs context window — three limits, enforced independently, and which fix follows from each.

This page counts anonymous readership with one counter, the file at /measure.js, and each of the events below is sent at most once per load. Once the page is ready it sends one view event. With it go the page path, the domain of the page you came from — only the domain, and nothing at all if you came from this site — how long the page has been open, counting only the time it was actually in front of you and not the time it sat in a background tab, whether you have scrolled, and any campaign or outreach code in the link you followed. A second event, which we call a human candidate, is sent only once the page has also received a real input event from you — a mouse movement, a touch, a scroll or a key press — and has been visible in the foreground for ten seconds in all; a program that fetches the page, or opens it and sits there, cannot produce one. A third, “engaged”, is stricter still: it is sent only after the human-candidate event, once the page has been in front of you for ninety seconds in all and you have scrolled far enough to reach a marker we put where the explaining part of this page ends; a reader who stops short of that marker never produces one. The second and third events carry the same fields as the first. If you switch away from the tab or close it, an event that has just become due may be sent as you leave. If this page has a button or a copy control that says it records a click, pressing it sends the name of that event and nothing else. Following a link to buy the guide sends one further event, also at most once per load, carrying that event’s name and a single word for which of the two checkouts you were sent to — the Gumroad listing or the payment page on this site — and nothing else: not the address you followed, not which page you were reading, not how long you had been there. Everything is sent to this site only, with no cookie attached. No cookie is read or written, nothing is put in your browser’s local or session storage, and no identifier is made from your device. Earlier versions of this page kept the campaign codes of your first visit in local storage under the name llab_attr; this page neither reads nor deletes that entry, so if it is there it stays until you clear this site’s data. An outreach code is minted per recipient, which would let us tell one reader from another; it is sent with each event and, unlike on earlier versions of this page, it is no longer removed from the address bar after it is read — if you copy the address, the code goes with it. Earlier versions also asked this domain for Vercel’s analytics script at /_vercel/insights/script.js, which on 26 September 2026 returned HTTP 404 on every host we publish; this page no longer asks for it. Your browser and the network attach things the page does not send: the identification string your browser gives, your IP address and the time of the request. The host that serves this page keeps its own request logs; those are its record and not ours.