langchain-ai/langchain · error · ValueError

Asked to cache, but no cache found at `langchain.cache`.

Error message

Asked to cache, but no cache found at `langchain.cache`.

What it means

`ValueError` raised in sync `_generate_with_cache` when the model was asked to cache (`cache=True` on the call or model) but no cache object is set — neither a `cache` argument, a model-level cache, nor the global cache at `langchain_core.globals.get_llm_cache()`. The library refuses to silently skip caching because the caller explicitly requested it.

Source

Thrown at libs/core/langchain_core/language_models/chat_models.py:1917

                        else msg
                    )
                    for msg in messages
                ]
                prompt = dumps(normalized_messages)
                cache_val = llm_cache.lookup(prompt, llm_string)
                if isinstance(cache_val, list):
                    converted_generations = self._convert_cached_generations(cache_val)
                    self._replay_v2_events_for_cache_hit(
                        converted_generations,
                        run_manager=run_manager,
                        **kwargs,
                    )
                    return ChatResult(generations=converted_generations)
            elif self.cache is None:
                pass
            else:
                msg = "Asked to cache, but no cache found at `langchain.cache`."
                raise ValueError(msg)

        # Apply the rate limiter after checking the cache, since
        # we usually don't want to rate limit cache lookups, but
        # we do want to rate limit API requests.
        if self.rate_limiter:
            self.rate_limiter.acquire(blocking=True)

        # v2 streaming: preferred over v1 when any attached handler opts in via
        # `_V2StreamingCallbackHandler`. Drives the protocol event generator
        # (native or `_stream` compat bridge) through the shared helper so
        # `on_stream_event` fires per event, then returns a normal `ChatResult`
        # so caching / `on_llm_end` stay on the existing generate path.
        if self._should_use_protocol_streaming(
            async_api=False,
            run_manager=run_manager,
            **kwargs,
        ):
            stream_accum = ChatModelStream(

View on GitHub (pinned to e32fa9a52e)

Solutions

1. Set a global cache once at startup: `from langchain_core.globals import set_llm_cache; from langchain_core.caches import InMemoryCache; set_llm_cache(InMemoryCache())`.
2. Or pass a concrete cache per call/model: `model.invoke(prompt, cache=SQLiteCache(...))` — no global needed.
3. Or set `cache=False` if caching is not actually wanted.
4. Note in newer versions the global lives in `langchain_core.globals` (not top-level `langchain.cache`); import from the right module.

Example fix

# before
model.invoke("hi", cache=True)  # ValueError: no cache configured

# after
from langchain_core.globals import set_llm_cache
from langchain_core.caches import InMemoryCache
set_llm_cache(InMemoryCache())
model.invoke("hi", cache=True)

from langchain_core.globals import get_llm_cache
if cache_flag and get_llm_cache() is None and cache_obj is None:
    from langchain_core.caches import InMemoryCache
    from langchain_core.globals import set_llm_cache
    set_llm_cache(InMemoryCache())

try:
    result = model.invoke(prompt, cache=True)
except ValueError as e:
    if "no cache found" in str(e):
        set_llm_cache(InMemoryCache())
        result = model.invoke(prompt, cache=True)
    else:
        raise

When it happens

Trigger: Calling `.invoke`/`.generate` with `cache=True` (or instantiating `BaseChatModel(cache=True)`) without ever calling `set_llm_cache(InMemoryCache())` (or another `BaseCache`) and without passing a `cache` object.

Common situations: Copy-pasting code that relies on a global cache set elsewhere (e.g. in a notebook that was restarted); enabling caching in one process but forgetting in workers; version upgrades where cache setup moved out of `get_chain` helpers.

Related errors

• maxsize must be greater than 0
• No global cache was configured. Use `set_llm_cache`.to set a
• Could not resolve content_key {full_path!r}: expected a mapp
• Could not resolve content_key {full_path!r}: missing key {ke
• Unsupported cache value {cache}

AI-assisted analysis of langchain-ai/langchain@e32fa9a52e (2026-08-14). Data as JSON: /api/errors/d7ace6b344393e8f. Report an issue: GitHub.