Known Issues

This section lists the bugs and limitations that Eulix has today, how serious they are, how we know about them, and what to do in the meantime.

Scope: all subsystems Last verified against: ad744e9 (2026-11-03); newer entries name their own commit

How to read this section

Every entry has an ID, a severity, a status and a verification note, then a short description of the symptom, the cause, a workaround and the direction of a fix. The IDs are stable, so other documents can link to them.

Severity is a suggestion for triage, not a formal score:

Level Meaning
High Wrong answers or wrong data in a common situation, or a security concern.
Medium Wrong or missing results in a specific situation, or a real cost to users.
Low Cosmetic, rare, or only affects developers.

Status:

Status Meaning
Open A bug. Not fixed.
Partly fixed Some part is fixed; the rest remains. The entry says which.
By design Behaviour that was chosen on purpose. Listed because it surprises people.
Limitation Something Eulix does not do, and is not planned to do soon.
Won’t fix A known bug we chose not to fix. The entry says why.

Verification says how confident to be in the entry:

Label Meaning
Reproduced I ran it against the commit the page names (2026-11-03 unless the entry gives another date) and saw the problem.
Read in code I found it by reading the current source; I did not run it.
Carried over From an earlier document written by the maintainers. I did not re-check it.

Anything fixed is not listed here as an open problem. Fixed items are kept for the record in Fixed Issues, so you can see what used to be broken and which test guards it.

Pages

Page Covers
Parser And PRISM Parsing gaps and call-graph resolution (libeulix, eulix-parser).
Retrieval And Query How context is chosen (eulix-cli retrieval and router).
Embeddings The embedding tool and the query embedder.
CLI, MCP And TUI The command line, the chat screens, the MCP server and the model client.
LSP The eulix-lsp language server: rename, URIs, indexing cost and memory.
Fixed Issues Resolved problems and the tests that guard them.

Summary of open items

ID Issue Severity Status Verification
PRISM-1 Call graphs are approximate Medium By design Carried over
PRISM-2 PRISM v2 never links calls to methods in Go, Java, JavaScript, C and C++ High Partly fixed Reproduced
PRISM-3 Calls on a receiver of unknown type get no edge in v2 Medium By design Reproduced
PRISM-4 Calls resolve across languages High Open Reproduced
PRISM-5 Go and Java receivers never reach the resolver; Java new calls never resolve Medium Fixed Reproduced
PRISM-6 JavaScript classes record no base class Medium Open Reproduced
PRISM-7 C++ calls are recorded twice Medium Open Reproduced
PRISM-8 PRISM v2 is very slow on large Rust workspaces with shared type names Medium Open Reproduced
PRISM-9 PHP call graph is partial for dynamic code Low Limitation Carried over
PRISM-10 Go calls link to same-named struct types and to look-alike functions High Fixed Reproduced
PRISM-11 JavaScript calls on built-in objects link to project functions of the same name Medium Open Reproduced
PRISM-12 TypeScript ignores declared types; new calls never link Medium Open Reproduced
PRISM-13 C++ member calls on library objects link to project methods of the same name High Open Reproduced
PARSE-1 Unsupported languages (Ruby, C# and others) Medium Limitation Read in code
PARSE-2 .jsx and .tsx files are rejected High Open Reproduced
PARSE-3 C and C++ files parse only partly Low By design Carried over
PARSE-4 dependency_graph is never filled; some Cargo features do nothing Low Open Read in code
PARSE-7 Java methods inside anonymous classes are not extracted Medium Open Reproduced
PARSE-8 Overloads and same-name definitions in one file share one id Medium Open Reproduced
PARSE-9 JavaScript functions that are not declared or bound to a name have no node Medium Open Reproduced
RET-1 The kb_exact search strategy never returns anything High Open Read in code
RET-2 Cross-root demotion only works for two folder names Medium Open Read in code
RET-3 Deep multi-hop questions may be answered inaccurately Medium Open Carried over
RET-4 EHNAS: word-feedback drift hurts some conceptual questions Medium Open Carried over
RET-5 EHNAS: several benchmark questions still fall back to text search Medium Open Carried over
RET-6 The Monte Carlo lexical layer depends on map order Low Open Carried over
RET-7 Embedder is never restarted after a timeout Medium Open Carried over
RET-8 Names from a stale knowledge base can still appear Low Open Carried over
RET-10 Rewriting kb.json while Eulix reads it can crash the reader Low Won’t fix Reproduced
RET-11 Quitting chat during a vector search can crash on Linux Low Won’t fix Read in code
RET-12 A knowledge-base file that fails to decode is parsed twice Low Won’t fix Read in code
RET-13 A question naming a Python method gets the class, not the method Medium Open Reproduced
EMB-1 ONNX on ROCm produces wrong embeddings High Partly fixed Carried over
EMB-2 A repository still chooses which embedding model is downloaded Medium Open Carried over
EMB-3 ROCm prints (null): No such file or directory Low Limitation Carried over
EMB-4 Profiling the launcher misses the embedding worker’s memory Low Limitation Carried over
EMB-5 Unused and unsafe Embedder methods Low Open Carried over
EMB-6 Embedded chunk text has no line breaks High Open Reproduced
EMB-7 Chunk text omits calls, complexity, exceptions, imports and docstrings Medium Open Reproduced
CLI-1 Each chat answer is probably saved to history twice Medium Open Read in code
CLI-2 The history viewer deletes the wrong entry, does not refresh and has no way back Medium Open Read in code
CLI-3 [checksum] settings are ignored Low Open Read in code
CLI-4 eulix query prints a prompt and never calls the model Low By design Read in code
CLI-5 Model requests cannot be cancelled, are not streamed, and have three smaller defects Medium Open Read in code
CLI-6 History entries with equal timestamps list in arbitrary order Low Open Read in code
CLI-7 Developer-facing loose ends (logger races, lint config, stale comments, dead repair tools) Low Open Read in code
CLI-11 “Where are X run” is answered with where X is defined Medium Open Reproduced
MCP-1 A repository’s own .eulix/ and eulix.toml are trusted High Open Carried over
MCP-2 A panic in a handler takes the MCP server down Low Open Read in code
MCP-3 MCP text-search fallback gaps Low Open Read in code
MCP-6 Project memory adds unrelated “Key symbols” to later queries Medium Open Reproduced
LSP-1 Rename edits names that the workspace does not define High Open Reproduced
LSP-2 Rename and go-to-definition misbehave with editor-encoded URIs High Open Reproduced
LSP-3 Watcher and reparse cost still grows with the callers of common names Medium Open Reproduced
LSP-4 References and incoming calls on common names are slow and memory-heavy Medium Open Reproduced
LSP-5 The index takes as much memory as the parsed files Medium Open Reproduced
LSP-6 The scan memory budget and lazy scan options do nothing Low Open Read in code
LSP-7 Parsing never uses more than 4 threads Low Open Reproduced
LSP-8 Shutdown waits for a running index commit Low Open Reproduced
LSP-9 C macros are not indexed (structs in headers fixed) Medium Partly fixed Reproduced
LSP-10 Hover shows nothing on calls into other files Low Limitation Reproduced