gmakstutis/graphiti
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
graphiti/ISSUE_TRIAGE.md
Daniel Chalef 31e8d60cc3 Convert all issue numbers to GitHub links 
- Converted bare issue numbers to proper markdown links
- Format: [#123](https://github.com/getzep/graphiti/issues/123)
- Applied to all three triage documents
- Makes issues clickable in GitHub web interface
2025-10-07 09:23:55 -07:00

16 KiB
Raw Blame History

GitHub Issues Triage & Categorization

Date: 2025-10-07 Total Open Issues: 100+ Analysis Scope: Issues #451 - #972

Issue Clusters

1. Database Driver Support

FalkorDB Issues (6 issues)

  • #972 - [BUG] Vector type mismatch in cosine distance operations

    • Component: component:database
    • Status: Open

  • #815 - [BUG] falkordb query

    • Component: component:database
    • Status: Open

  • #757 - [BUG] run quickstart_falkordb example, get query error

    • Component: component:database
    • Status: Open

  • #749 - [BUG] Official Docker Image doesn't support FalkorDB

    • Component: component:database, component:mcp
    • Status: Open

  • #731 - Episode insertion fails due to malformed Cypher query

    • Component: component:database
    • Status: Open

  • #719 - FalkorDB support in MCP-Server

    • Component: component:mcp, component:database
    • Status: Open

New Database Support Requests (6 issues)

  • #947 - Apache AGE Graph DB support
  • #933 - Support RDF
  • #781 - More graph database support (NebulaGraph mentioned)
  • #779 - Postgres with pgvector support
  • #644 - AWS Neptune driver
  • #643 - Kuzu driver support (may already be implemented?)
  • #642 - MemGraph driver

Analysis: FalkorDB has multiple critical bugs affecting basic operations. Consider deprecating or fixing comprehensively. High demand for alternative backends suggests need for driver abstraction layer.

2. MCP Server Issues (9 issues)

  • #923 - [BUG] Switch from SSE (deprecated) to Streamable HTTP

    • Priority: HIGH - using deprecated protocol
    • Component: component:mcp

  • #945 - [BUG] Custom OPENAI_BASE_URL causes NaN embeddings

    • Component: component:mcp, component:llm

  • #848 - [BUG] clear_graph tool fails silently (async session bug)

    • Component: component:mcp

  • #840 - [BUG] Failed to validate request (initialization timing)

    • Component: component:mcp

  • #723 - DEFAULT_MAX_TOKENS 8192 too restrictive

    • Component: component:mcp

  • #578 - MCP server with Gemini model

    • Component: component:mcp, component:llm

  • #565 - Cross-encoder ignores OPENAI_BASE_URL

    • Component: component:mcp, component:llm

  • #509 - MCP server add_nodes tool call doesn't work

    • Component: component:mcp

Duplicates in this cluster:

  • #867 + #831 - GPT-oss:20 and 120B models (duplicate)
  • #787 - Rate limit even with SEMAPHORE_LIMIT=1 (marked duplicate)

Analysis: MCP server has protocol deprecation issue (#923) and multiple configuration/provider issues. Should be high priority for users relying on MCP integration.

3. LLM Provider Compatibility (8 issues)

  • #902 - [BUG] OpenAI internal call broken (reasoning.effort param)

    • Component: component:llm

  • #878 - [BUG] GPT-5 temperature parameter unsupported

    • Component: component:llm

  • #912 - [BUG] Pydantic validation error with deepseek-r1:7b

    • Component: component:llm

  • #791 - [BUG] Small model setting defaults to gpt-4.1-nano

    • Component: component:llm

  • #790 - [BUG] Failed to parse structured response with Gemini

    • Component: component:llm

  • #868 - [BUG] Cannot work with Ollama

    • Component: component:llm

  • #763 - [BUG] LLMConfig.max_tokens not respected

    • Component: component:llm

  • #760 - [BUG] Hallucinations with default models

    • Component: component:llm

Additional LLM Provider Requests:

  • #907 - VSC Copilot models
  • #751 - Alibaba Cloud AI model
  • #724 - Gemini with Google Cloud Credentials
  • #459 - Amazon Bedrock support

Analysis: Structured output compatibility is a recurring theme. Many issues stem from providers not supporting OpenAI's structured output format. Documentation should clearly state which providers are fully compatible.

4. Duplicate Entities (3 issues)

  • #963 - [BUG] Duplicate entities in Neo4j (marked duplicate)

    • Component: component:core
    • Status: Duplicate (see #875)

  • #875 - [BUG] Duplicate entities with custom db name

    • Component: component:core, component:database
    • Priority: HIGH - core deduplication functionality broken

  • #774 - Same Chinese text extracted as different facts

    • Component: component:core

Analysis: Core deduplication functionality appears broken, especially with custom database names. Critical for production use.

5. Bulk Upload Issues (3 issues)

  • #882 - [BUG] IndexError during node resolution

    • Component: component:bulk

  • #879 - ValidationError 'duplicates' field missing

    • Component: component:bulk

  • #871 - Invalid JSON and index errors

    • Component: component:bulk

Related:

  • #747 - Add progress reporting to bulk upload
  • #658 - Bulk ingestion not possible (may be duplicate)

Analysis: Bulk operations are fundamentally broken. Appears to be schema/validation issues in the bulk processing pipeline.

6. Search Issues (6 issues)

  • #810 - [BUG] Empty group_id handled inconsistently in search

    • Component: component:search

  • #801 - [BUG] episode_fulltext_search empty results (marked duplicate)

    • Component: component:search

  • #838 - Allow searching across all groups when group_ids is None

    • Component: component:search

  • #772 - [BUG] BFS max_depth parameter ignored

    • Component: component:search

  • #789 - [BUG] BFS returns duplicate edges with swapped source/target

    • Component: component:search

  • #777 - [BUG] MMR reranker RuntimeWarning, no results

    • Component: component:search

Related:

  • #488 - edge_search_filter_query_constructor creating incorrect query
  • #534 - retrieve_episodes always returns no results

Analysis: Search functionality has multiple bugs in filtering, BFS traversal, and reranking. Group ID handling is particularly problematic.

7. Database Configuration (4 issues)

  • #851 - [BUG] Search only connects to 'neo4j' db, add_episode works with any name

    • Component: component:database
    • Priority: HIGH - inconsistent behavior

  • #798 - [BUG] Database name not passed through Graphiti object

    • Component: component:database
    • Priority: HIGH - related to #851

  • #715 - [Feature] Configure Neo4j database name (multi-DB support)

    • Component: component:database

Analysis: Database name configuration is broken. Hardcoded 'neo4j' default causes issues. Related to commit mentioned in CLAUDE.md about hardcoded database names.

8. Datetime/Timezone Issues (4 issues)

  • #920 - [BUG] edge_operations.py timezone-naive/aware comparison (marked duplicate)

    • Component: component:core
    • Status: Duplicate (possibly fixed in commit 73015e9)

  • #893 - [BUG] Kuzu driver valid_at datetime format

    • Component: component:database

  • #606 - Add support for datetime fields in custom entities

    • Component: component:core

Analysis: Recent commit 73015e9 "Fix datetime comparison errors by normalizing to UTC" may have addressed some of these. Needs verification.

9. API/Server Issues (4 issues)

  • #961 - [Feature Request] Improve Episodes API (UUID, GET by ID, metadata)

    • Component: component:server

  • #921 - SDK client like zep-cloud/zep-python

    • Component: component:server

  • #566 - /messages endpoint doesn't persist episodes

    • Component: component:server

  • #904 - docker-compose.yml env vars override .env

    • Component: component:server, component:mcp

Analysis: API functionality gaps and configuration issues. Episode persistence bug is critical.

10. Error Handling & Validation (3 issues)

  • #941 - TaskGroup errors (marked duplicate)

    • Component: component:core

  • #937 - ArgumentError with empty query entities

    • Component: component:core

  • #951 - Incorrect try import for AsyncOpenSearch

    • Component: component:database

Analysis: Input validation needs improvement to handle edge cases like empty strings.

11. Feature Requests - Core Functionality (10 issues)

  • #935 - Create episodes based on DOM structure
  • #934 - Flag contradictions on merging facts
  • #925 - Monitor LLM conversations
  • #905 - ColbertV2 embeddings with Fastembed
  • #864 - How to forget knowledge
  • #819 - Count token usage
  • #669 - Metadata on chunks for RAG
  • #465 - Support ignoring non-custom entities
  • #467 - LLM inference expenses are high

Analysis: Feature requests range from observability (#925, #819) to advanced RAG features (#669, #905). Token cost reduction (#467) is recurring concern.

12. Documentation/Questions (9 issues)

  • #913 - How to update data in graph
  • #909 - Reproduce LongMemEval results
  • #869 - Portuguese: Change OpenAI model
  • #853 - Document VCS/extension setup
  • #828 - Get a flowchart
  • #701 - Chinese video tutorial (informational)
  • #484 - OpenAIGenericClient documentation
  • #530 - Cursor AI + OpenRouter setup
  • #517 - OpenRouter and Voyage setup

Component: component:docs

Analysis: Many questions indicate documentation gaps, especially around custom LLM provider setup and basic operations.

13. Code Quality/Refactoring (5 issues)

  • #836 - [BUG] update_communities broken

    • Component: component:core

  • #811 - Refactor: hoist EXTRACT_EDGES_MAX_TOKENS constant

    • Component: component:core

  • #681 - Refactor: no internal ghost variables

    • Component: component:core

  • #451 - Minor type bug in bulk_utils

    • Component: component:bulk

  • #717 - Adopt uv workspace for monorepo

    • Component: component:core, component:server, component:mcp

Analysis: Technical debt items. #717 (uv workspace) would improve developer experience for monorepo.

14. Cloud/Alternative Providers (2 issues)

  • #739 - MCP server based on zep-cloud

    • Component: component:mcp

  • #538 - Azure Cosmos DB version (informational)

15. Embeddings/Reranking (3 issues)

  • #728 - Voyage embedder installation issue

    • Component: component:core

  • #485 - Better Embedder error information

    • Component: component:core

  • #543 - Hard-coded model in OpenAIRerankerClient

    • Component: component:core

16. Miscellaneous Bugs (5 issues)

  • #800 - Blank disconnected nodes in sample project

    • Component: component:mcp

  • #686 - Relationships not extracted automatically

    • Component: component:core

  • #687 - Unable to add episodes with OpenAI agents SDK

    • Component: component:core

  • #587 - Node type not being set

    • Component: component:core

Summary Statistics

  • Total Clustered Issues: 100+
  • Major Clusters: 16
  • Confirmed Duplicates: 6 (marked)
  • High Priority Bugs: ~15-20
  • Feature Requests: ~20
  • Documentation Gaps: ~10

Component Breakdown

  • component:core - 25+ issues
  • component:database - 20+ issues (includes all DB drivers)
  • component:mcp - 12+ issues
  • component:llm - 12+ issues
  • component:search - 8+ issues
  • component:bulk - 5+ issues
  • component:server - 4+ issues
  • component:docs - 10+ issues