Problemsby GlyphLlamaIndex upgrades break working setups
17 posts from 10 people7d activeFading, +0 this week
Posts per day
The posts behind it
[Bug]: SemanticDoubleMergingSplitterNodeParser ignores include_prev_next_rel=False
### Bug description `SemanticDoubleMergingSplitterNodeParser.from_defaults(include_prev_next_rel=False, ...)` still attaches `PREVIOUS` and `NEXT` relationships. Its `build_semantic_nodes_from_nodes()` links the chunks unconditionally before the base parser's postprocessing...
[Bug]: TransformRetriever.aretrieve invokes the synchronous backend and transform
### Bug description `TransformRetriever` implements `_retrieve()` but has no `_aretrieve()`. Consequently, `aretrieve()` inherits the base fallback, runs the synchronous query transform on the event loop, and calls the wrapped retriever's `retrieve()` rather than `aretrieve()`....
[Bug]: partial_format can drop the original output parser or fail to copy chat prompts
### Bug description `PromptTemplate.partial_format()` temporarily sets `self.output_parser = None` while deep-copying the prompt. If copying any other prompt state raises, the original template is permanently left without its parser, so later formatting silently loses its output...
[Bug]: SimplePropertyGraphStore relation upserts ignore updated properties
### Bug description Calling `SimplePropertyGraphStore.upsert_relations()` with an existing `(source, label, target)` silently ignores updated relation properties. `LabelledPropertyGraph.add_triplet()` returns early for an existing triplet before replacing the relation object....
[Bug]: TextToCypherRetriever drops the user's question during response summarization
### Bug description With `summarize_response=True`, both `retrieve_from_graph()` and `aretrieve_from_graph()` pass `question=parsed_cypher_query` to the summarization LLM. The summarizer receives the generated Cypher in the question slot instead of the original natural-language...
[Bug]: SimpleDirectoryReader fails with fsspec DirFileSystem: 'PurePosixPath' object is not iterable
### Bug Description When SimpleDirectoryReader is given an fsspec DirFileSystem, load_data() fails with TypeError: 'PurePosixPath' object is not iterable. load_file() passes a PurePosixPath to fs.open(). DirFileSystem._join() only handles str or list inputs, so it tries to...
[Bug]: `PickleSerializer` returns a different value for some JSON-serialized values, e.g. the string `"SwEu"` comes back as `1`
### Bug Description `PickleSerializer` serializes with JSON when it can and falls back to pickle, but `deserialize` tries pickle first: `base64.b64decode` without `validate=True` silently drops every character outside the base64 alphabet, so a JSON payload is also decoded as...
fix(vector_stores): SimpleVectorStore safe deserialization, ref_doc_id collision, and metadata key isolation
...- Calling `persist("store.json")` passes `dirpath=""` to `fs.makedirs("")`, failing on strict filesystems. - Serializing JSON payload prior to opening the target path with `"w"` prevents truncating an existing store to `0` bytes if serialization raises `TypeError`. #### Proposed...
[Bug]: `PGRetriever` never calls `BaseRetriever.__init__`; `callback_manager` passed to it is dropped and `object_map` is unset
### Bug Description `PGRetriever.__init__` sets its own attributes but never calls `BaseRetriever.__init__`, so `callback_manager`, `object_map` and `_verbose` are never initialized on the instance. This has been the case since the class was added. Two symptoms: 1. A...
SimpleDirectoryReader required_exts is case-sensitive
### Bug Description `SimpleDirectoryReader(required_exts=[".txt"])` skips files with an uppercase extension such as `REPORT.TXT`, `scan.PDF` or `photo.JPG`. The same reader picks a file reader by `input_file.suffix.lower()` in `load_file` and `aload_file`, so these files would...
[Bug]: DirectLookaheadAnswerInserter splices answers at stale offsets when a lookahead has multiple [Search()] tags
### Bug Description `DirectLookaheadAnswerInserter.insert` walks `zip(query_tasks, answers)` and rebuilds the response string for each task: The `start_idx`/`end_idx` on each `QueryTask` point into the original response string (that is where `QueryTaskOutputParser` measured...
[Bug]: HTMLNodeParser rips inline tag text out of its sentence and emits it as a separate node
### Bug Description `HTMLNodeParser`'s default tag list includes the inline formatting tags `b`, `i`, and `u`. When one of these appears inside a block tag, the parser drops the inline tag's text from the surrounding sentence and emits that text as its own node after the block....
[Bug]: BaseWorkflowAgent.init_run raises IndexError when chat_history contains no user messages
### Bug Description In `llama_index.core.agent.workflow.base_agent.BaseWorkflowAgent.init_run`, when initializing an agent workflow with a `chat_history` that contains non-system messages but no `user` messages (e.g., resuming from an initial assistant greeting or customer...
[Bug]: RichPromptTemplate crashes on embedded audio/video files and corrupts embedded documents
...URL, relative URL without a base`. Any chat template embedding an audio or video file is unusable. - **Document files are silently corrupted.** `DocumentBlock.url` is a plain string, so the base64 blob "succeeds" and every downstream consumer receives a document block whose...
Ingestion cache contract: stale node identity vs content-stable reuse after regenerated chunk IDs
## Context PR #23003 fixes the document-level identity collision from #23002, but its follow-up discussion exposed a broader cache-contract choice that appears worth deciding separately rather than burying in that focused fix. The current generic transformation cache stores and...
[Bug]: SimplePropertyGraphStore.get_rel_map returns duplicate triplets when the graph has cycles
### Bug Description `SimplePropertyGraphStore.get_rel_map` walks the graph depth by depth and tracks visited triplets in `seen_triplets`, but the seen-set starts empty and is only updated with the *next* depth's triplets: Any path that loops back to an already-visited depth - a...
McpToolSpec exposes dynamic MCP resource templates as zero-argument read_resource tools
...a URI template such as: The generated FunctionTool has no arguments and therefore has no way to bind `owner`/`name` before calling `read_resource()`. It attempts to read the template URI literally. ## Minimal Reproduction Shape A fake MCP client is sufficient: Then: The...
Companies and products named
About this problem
Evidence
17 posts from 10 people in 1 place, about 2 a week over 46 days. Mostly on run-llama/llama_index. Tools named alongside: Model Context Protocol.
Frustration Frustration 1 of 3· Seen on GitHub issues
Summary, model-written
LlamaIndex users encounter unusable chat templates, corrupted document files, incorrect response rebuilding, and graph traversal loops after recent code updates.
How it was grouped
Posts that state a pain and match the "breaking-changes" rule about LlamaIndex. First post Sep 8, 2026, latest Oct 23, 2026. Corroborated: 3 or more posts from 2 or more people or places. Method
The same problem elsewhere
GitHub upgrades break working setups
87 posts from 75 people on GitHub issues, Hacker News, Stack Overflow.
Upgrades and new releases break working setups
76 posts from 62 people on GitHub issues, Reddit, App Store reviews.
Claude Code upgrades break working setups
67 posts from 60 people on GitHub issues, Hacker News, Reddit.
Codex upgrades break working setups
65 posts from 58 people on GitHub issues, Reddit.
DuckDB upgrades break working setups
37 posts from 28 people on GitHub issues.
Claude upgrades break working setups
35 posts from 34 people on GitHub issues, Reddit, Hacker News.
ClickHouse upgrades break working setups
31 posts from 20 people on GitHub issues.
Anthropic upgrades break working setups
27 posts from 22 people on GitHub issues.
PyTorch upgrades break working setups
26 posts from 21 people on GitHub issues, Reddit.
ChatGPT upgrades break working setups
26 posts from 26 people on GitHub issues, Community forums, App Store reviews.
OpenAI upgrades break working setups
25 posts from 25 people on GitHub issues, Reddit, Community forums.
Open WebUI upgrades break working setups
25 posts from 2 people on GitHub issues.
Other LlamaIndex problems
RAG on LlamaIndex returns irrelevant or missing context
5 posts from 5 people on GitHub issues.
LlamaIndex users keep filing bugs
Top post: "SQLAlchemyChatStore.delete_message treats positional index as database primary key", one of 17 posts from 12 people on GitHub issues, Stack Overflow.
LlamaIndex fails to return valid structured output
5 posts from 5 people on GitHub issues.
Trouble with LlamaIndex returns
Top post: "LegacyOfficeReader returns a Document with text None for a document with no text", one of 3 posts from 3 people on GitHub issues.
LlamaIndex vector search quality is hard to get right
3 posts from 3 people on GitHub issues.
Trouble with LlamaIndex functiontool
Top post: "FunctionTool wraps a returned ToolOutput as a success and drops is_error", one of 3 posts from 3 people on GitHub issues.
Trouble with LlamaIndex documents
Top post: "MoorchehVectorStore sparse-vector path raises NameError (missing imports, undefined 'documents', ImpotError", one of 3 posts from 3 people on GitHub issues.
Trouble with LlamaIndex default
Top post: "max_iterations=0 is silently replaced by the default 20", one of 3 posts from 3 people on GitHub issues.
Experiments on LlamaIndex are hard to reproduce and track
3 posts from 3 people on GitHub issues.
LlamaIndex lock-in makes switching costly
3 posts from 3 people on GitHub issues.
Trouble with LlamaIndex events
Top post: "CBEvent assigns generated uuid to self.id instead of self.id_, so events without an id all share id_=", one of 3 posts from 3 people on GitHub issues.
LlamaIndex integrations break or are missing
2 posts from 2 people on GitHub issues. Unverified.
History