infiniflow/ragflow · error · Exception

No memory is selected.

Error message

No memory is selected.

What it means

The Retrieval component's memory path (_retrieve_memory, agent/tools/retrieval.py:272) collects self._param.memory_ids and calls MemoryService.get_by_ids. If the lookup returns no rows it raises Exception('No memory is selected.'), i.e. the component is configured to search memories but no configured memory ID resolves to an existing record.

Source

Thrown at agent/tools/retrieval.py:272

        # Format the chunks for JSON output (similar to how other tools do it)
        json_output = kbinfos["chunks"].copy()

        self._canvas.add_reference(kbinfos["chunks"], kbinfos["doc_aggs"])
        form_cnt = "\n".join(kb_prompt(kbinfos, 200000, True))

        # Set both formalized content and JSON output
        self.set_output("formalized_content", form_cnt)
        self.set_output("json", json_output)

        return form_cnt

    async def _retrieve_memory(self, query_text: str):
        memory_ids: list[str] = [memory_id for memory_id in self._param.memory_ids]
        user_id: str = self._param.user_id if hasattr(self._param, "user_id") else None
        memory_list = MemoryService.get_by_ids(memory_ids)
        if not memory_list:
            raise Exception("No memory is selected.")

        embd_names = list({memory.embd_id for memory in memory_list})
        assert len(embd_names) == 1, "Memory use different embedding models."

        vars = self.get_input_elements_from_text(query_text)
        vars = {k: o["value"] for k, o in vars.items()}
        query = self.string_format(query_text, vars)
        # query message
        filter_dict: dict = {"memory_id": memory_ids}
        if user_id:
            import re

            # is variable
            if re.match(r"^{.*}$", user_id):
                user_id = self._canvas.get_variable_value(user_id)
            filter_dict["user_id"] = user_id
        message_list = memory_message_service.query_message(
            filter_dict, {"query": query, "similarity_threshold": self._param.similarity_threshold, "keywords_similarity_weight": self._param.keywords_similarity_weight, "top_n": self._param.top_n}

View on GitHub (pinned to 554fb1133a)

Solutions

1. Open the Retrieval component config and select at least one existing memory in the memory_ids picker.
2. Verify the memory still exists (check the memory/dataset admin page) and re-select it if it was recreated with a new ID.
3. If memory search is optional, disable the memory option instead of leaving it on with an empty selection.
4. When importing agents from another deployment, rebind memory IDs to local ones.

Example fix

// before
"memory_ids": []  // memory search enabled but nothing selected

// after
"memory_ids": ["mem_9c3d..."]

from api.db.services.memory_service import MemoryService

if memory_search_enabled:
    if not memory_ids or not MemoryService.get_by_ids(list(memory_ids)):
        raise ValueError("Memory search is enabled but no valid memory is selected")

try:
    await retrieval._retrieve_memory(q)
except Exception as e:
    if "No memory is selected" in str(e):
        # degrade to plain retrieval instead of failing the whole canvas
        return await retrieval._retrieve(q)
    raise

When it happens

Trigger: memory_ids is empty on a Retrieval component whose mode/inputs route it into _retrieve_memory; the configured memory records were deleted; memory IDs copied from another tenant/environment that do not exist locally.

Common situations: Agent template references a memory that was removed; user toggles 'use memory' but never picks a memory; importing agent JSON across environments.

Related errors

• No dataset is selected.
• Dataset({nm_or_id}) does not exist.
• SANDBOX_LOCAL_MAX_MEMORY_MB must be greater than 0.
• Invalid response from SearXNG
• Webhook security is required. Set allow_anonymous to true to

AI-assisted analysis of infiniflow/ragflow@554fb1133a (2026-08-15). Data as JSON: /api/errors/595415bc5438e17f. Report an issue: GitHub.