gmakstutis/graphiti
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
graphiti/PRIORITY_ROADMAP.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

12 KiB
Raw Blame History

Priority Roadmap - GitHub Issues

Date: 2025-10-07 Purpose: Identify and prioritize critical issues requiring immediate attention

P0 - Critical (Fix Immediately)

1. MCP Server Protocol Deprecation

Issue #923 - Switch from SSE to Streamable HTTP

  • Impact: Using deprecated MCP protocol that will be removed
  • Affected Users: All MCP server users
  • Risk: Protocol will stop working when support is removed
  • Effort: Medium
  • Component: component:mcp

Action: Migrate to new protocol ASAP

2. Database Name Configuration Broken

Issues #851, #798 - Database name not passed/respected

  • Impact: Cannot use non-default Neo4j databases for search operations
  • Affected Users: Multi-database deployments, production environments
  • Root Cause: Hardcoded 'neo4j' database name in search operations
  • Symptoms:
    • Search only works with default 'neo4j' database
    • add_episode works with any database name (inconsistent)
  • Effort: Low-Medium
  • Component: component:database, component:core

Action: Fix database name propagation through entire codebase

3. Duplicate Entity Detection Broken

Issue #875 (duplicate: #963)

  • Impact: Core deduplication functionality not working
  • Affected Users: All users, especially with custom database names
  • Symptoms: Same entities created multiple times instead of being merged
  • User Reports: Tested with multiple LLM models, issue persists
  • Effort: Medium-High
  • Component: component:core

Action: Debug entity resolution and deduplication logic

P1 - High Priority (Fix This Sprint)

4. Bulk Upload Completely Broken

Issues #882, #879, #871 - Multiple bulk upload failures

  • Impact: Bulk operations unusable
  • Affected Users: Anyone trying to import large datasets
  • Symptoms:
    • IndexError during node resolution (#882)
    • ValidationError for missing 'duplicates' field (#879)
    • Invalid JSON errors (#871)
  • Effort: Medium
  • Component: component:bulk

Action: Fix validation schema and error handling in bulk pipeline

5. Datetime Comparison Issues

Issues #920, #893, #606

  • Impact: Crashes when comparing dates
  • Status: May be partially fixed by commit 73015e9
  • Affected Users: All users with temporal queries
  • Effort: Low (if recent fix is complete)
  • Component: component:core, component:database

Action: Verify recent datetime UTC normalization fix covers all cases

6. Search Group ID Handling

Issues #810, #838, #801 - Inconsistent group_id behavior

  • Impact: Search returns empty results or behaves inconsistently
  • Affected Users: Multi-tenant deployments using group_ids
  • Symptoms:
    • Empty group_id handled differently in fulltext vs similarity search
    • Cannot search across all groups
  • Effort: Low-Medium
  • Component: component:search

Action: Standardize group_id handling across search methods

7. BFS Search Bugs

Issues #772, #789 - BFS traversal broken

  • Impact: Graph traversal doesn't work as expected
  • Symptoms:
    • max_depth parameter completely ignored (#772)
    • Duplicate edges with swapped source/target (#789)
  • Effort: Medium
  • Component: component:search

Action: Fix BFS implementation

P2 - Medium Priority (Address Soon)

8. MCP Server Configuration Issues

Issues #945, #840, #848, #565 - Various MCP bugs

  • #945: Custom OPENAI_BASE_URL causes NaN embeddings
  • #840: "Failed to validate request" initialization timing
  • #848: clear_graph fails silently (async session bug)
  • #565: Cross-encoder ignores OPENAI_BASE_URL

Impact: MCP server unreliable with custom configurations Effort: Low-Medium per issue Component: component:mcp

9. LLM Provider Compatibility

Issues #878, #902, #912, #791 - Provider-specific bugs

  • #902: OpenAI reasoning.effort parameter breaks API
  • #878: GPT-5 doesn't support temperature parameter
  • #791: Small model setting ignored, always defaults to gpt-4.1-nano
  • #912: Pydantic validation errors with non-OpenAI models

Impact: Specific LLM models/providers broken Effort: Low per issue Component: component:llm

Strategy: Add provider capability detection and parameter filtering

10. FalkorDB Driver Issues

Issues #972, #815, #757, #731, #749 - FalkorDB broken

  • Impact: FalkorDB backend largely non-functional
  • Effort: High (multiple issues)
  • Component: component:database

Decision Point: Fix comprehensively or deprecate? Appears unmaintained.

11. Error Handling Improvements

Issues #937, #951 - Edge case crashes

  • #937: Empty query strings cause ArgumentError
  • #951: Incorrect import fallback for AsyncOpenSearch

Impact: Crashes on edge cases Effort: Low per issue Component: component:core

12. API Server Issues

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

  • Impact: Core API functionality broken
  • Effort: Medium
  • Component: component:server

Action: Debug why episodes aren't being persisted

13. Code Quality Issues

Issues #836, #811, #681, #451 - Technical debt

  • #836: update_communities broken (tuple unpacking error)
  • #811: Hoist hardcoded token constant
  • #681: Remove ghost variables
  • #451: Type bug in bulk_utils

Impact: Medium (functionality broken in #836, others are maintenance) Effort: Low per issue

P3 - Lower Priority / Feature Requests

14. Feature Enhancements

  • #961 - Improve Episodes API (UUID, GET by ID, metadata)
  • #935 - Create episodes based on DOM
  • #934 - Flag contradictions on merging facts
  • #925 - Monitor LLM conversations
  • #819 - Count token usage
  • #747 - Progress reporting for bulk upload
  • #669 - Metadata on chunks for RAG
  • #465 - Support ignoring non-custom entities

15. New Provider Support

  • #907 - VSC Copilot models
  • #905 - ColbertV2 embeddings
  • #751 - Alibaba Cloud AI
  • #724 - Gemini with GCP credentials
  • #459 - Amazon Bedrock
  • #739 - zep-cloud based MCP

16. Documentation Improvements

  • #913 - How to update data in graph
  • #853 - VCS/extension setup
  • #828 - Flowchart documentation
  • #484 - OpenAIGenericClient documentation
  • All "How to" questions

17. Alternative Database Support

Immediate Action Plan (Next 2 Weeks)

Week 1

  1. Fix #923 - MCP protocol migration (CRITICAL)
  2. Fix #851/#798 - Database name configuration
  3. Fix #875 - Duplicate entity detection
  4. Verify #920 - Datetime issues resolved by recent commit

Week 2

  1. Fix #882/#879/#871 - Bulk upload pipeline
  2. Fix #810 - Group ID handling in search
  3. Fix #772/#789 - BFS search issues
  4. Close duplicates - Process 6+ confirmed duplicates

Ongoing

  • Investigate duplicate clusters - Consolidate related issues
  • Document workarounds - For issues that can't be fixed immediately
  • Provider compatibility matrix - Document which LLMs work fully

Impact Summary

Critical Issues Blocking Production Use:

  • Database name configuration (multi-DB deployments)
  • Duplicate entity detection (core functionality)
  • Bulk upload failures (data ingestion at scale)
  • MCP protocol deprecation (future compatibility)

High-Impact Bugs Affecting Many Users:

  • Search group_id handling
  • BFS traversal issues
  • Datetime comparison problems
  • Various MCP server configuration bugs

Lower Impact:

  • Specific LLM provider incompatibilities (workarounds available)
  • FalkorDB issues (alternative: Neo4j)
  • Feature requests and enhancements

Component Labels to Apply

Use these labels when triaging the issues above:

  • component:core - Core library
  • component:mcp - MCP server
  • component:database - Database drivers
  • component:search - Search functionality
  • component:bulk - Bulk operations
  • component:server - FastAPI server
  • component:llm - LLM providers
  • component:docs - Documentation