Skip to content
Agents tracked: 501 Releases (7d): 777 Agent status: no reported issuesSince last limit reset: Claude 10d · Codex 8d For AI agents: Markdown Updated Nov 21, 2026

Graphiti changelog: what's new each month

Every stable Graphiti release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 12 months covered; the current month updates daily.

October 2026

3 releases: 0.31.0 → 0.31.2

October brought upgrades to MCP SDK 2.x, improved security defaults for the reference servers, and several fixes to core graph operations and search functionality.

Highlights

  • MCP server now upgraded to SDK 2.x for better compatibility.
  • Reference MCP and REST servers now listen only on localhost by default for improved security.
  • Fixed fulltext search to properly apply group filters.
  • Core graph operations now correctly handle fact restatement and edge recording.

New

  • Publish Graphiti on version tags via MCP, similar to other MCP projects.
  • Contributor guidelines and intake bot for issues and pull requests.

Improved

  • MCP tools now route to the graph of the requested group_id.
  • Documentation now clarifies the reference-only status and security scope of MCP and REST servers.

Fixed

  • Fulltext search now applies group filters correctly.
  • remove_episode now keeps restated facts and rebuilds summaries properly.
  • add_episode_bulk now records supported edges on episodes.
  • Reference servers listen only on localhost by default instead of all interfaces.

September 2026

3 releases: 0.30.0 → 0.30.2

September brought improvements to edge processing, Neo4j database routing, and data isolation in multi-group scenarios. The framework was reframed as a temporal knowledge graph solution with updated documentation.

Highlights

  • Added source and target node UUIDs and episodes to edge processing results for better traceability.
  • Fixed Neo4j database routing to respect configured database settings in all contexts.
  • Improved data clearing to include Saga nodes when removing data by group IDs.
  • Enhanced concurrent multi-group isolation with request-scoped driver management.
  • Fixed edge cross-encoder shortlist via balanced merge for more reliable ranking.

New

  • Source node UUID, target node UUID, and episodes now included in FactResult model.
  • NEO4JDATABASE environment variable now honored in MCP server.

Improved

  • Edge cross-encoder shortlist generation now uses balanced merge for better results.
  • Node attributes are now preserved when no entity type applies.
  • Multi-group decorator now binds arguments by name to prevent binding errors.
  • FalkorDB edge full-text search optimized to avoid full entity scans per hit.
  • Concurrent multi-group queries use request-scoped drivers for proper isolation.

Fixed

  • Neo4j queries now route to the configured database instead of default.
  • Data clearing by group IDs now includes Saga nodes.
  • Multi-group decorator argument binding fixed for correct parameter passing.
  • FalkorDB performance issue in edge full-text search resolved.

July 2026

1 release: 0.29.3

July brought dependency updates across the Graphiti ecosystem and improvements to testing infrastructure. The month focused on maintaining compatibility with newer versions of key libraries and strengthening reliability checks.

Highlights

  • Updated OpenAI library to version 2.41.0 and Google GenAI to 2.8.0 for latest API compatibility.
  • Added live end-to-end regression testing to catch issues before release.
  • Improved CI workflow reliability with enhanced error handling for edge cases.

New

  • Live end-to-end regression testing in CI pipeline to validate releases.
  • PR check integration with live testing workflows.

Improved

  • Enhanced error handling in live-test workflows with improved exit code diagnostics.
  • Updated Redis compatibility to support versions up to 9.
  • Updated Azure Identity library to 1.25.3 for improved authentication support.
  • Updated pytest-asyncio for better async test execution.

Fixed

  • Hardened exit-5 handling in CI workflows with better diagnostics.
  • Improved MCP backport handling in live-test workflows.

June 2026

1 release: 0.29.2

June brought stability improvements and expanded database support to Graphiti. The release focused on fixing compatibility issues and adding FalkorDB Lite as an embedded alternative.

Highlights

  • Added support for FalkorDB Lite, an embedded version of FalkorDB for simpler deployments.
  • Fixed Ollama and OpenAI-compatible provider support to work correctly.
  • Resolved Docker Compose configuration issues for FalkorDB.
  • Extended model environment variable overrides across all provider configurations.

New

  • FalkorDB Lite (embedded) support for lighter-weight graph database deployments.

Improved

  • Model environment overrides now apply consistently to all provider configurations, including routing.
  • Docker Compose alignment for FalkorDB service ports.
  • Provider routing and configuration handling for MCP.

Fixed

  • FalkorDB profile failing to start in Docker Compose.
  • Ollama and OpenAI-compatible provider support not working correctly.
  • Null bytes being retained in parameters.
  • Dependency vulnerabilities in uv.lock files.
  • FalkorDB Lite embedded installation issues (Redis version pinning).

May 2026

1 release: 0.29.1

May focused on strengthening extraction rules and improving recall across entities and facts. Rule expansions, new negative examples, and metric improvements reflect a month of refinement in the knowledge graph extraction pipeline.

Highlights

  • Entity and fact extraction rules expanded with sub-rules and new guidance for better coverage.
  • Negative-examples documentation added to clarify what should not be extracted by class.
  • Edge volume increased 7.7% with improved fact-text matching quality.
  • Precision and recall adjustments across entity types, with notable reductions in false positive classes.
  • Watermark filtering added to track which episodes have been processed.

New

  • ENTITY RULE 9 for didactic and tutorial scaffolding.
  • FACT RULE 8 for fact extraction.
  • Negative-examples block (A–G) with paired source-and-extraction guidance per class.
  • lastsummarizedat wall-clock filter for episode processing watermark.

Improved

  • ENTITY RULE 6 expanded from one rule into eight sub-bullets covering skip classes.
  • FACT RULE 4 strengthened for narrative announcements, forward commitments, plans, and emotional setup in conversational episodes.
  • Fact-text Jaccard similarity improved from 0.341 to 0.381 on matched edges.
  • Gold-edge recall increased by 0.9 percentage points.

Fixed

  • Quantities false positives reduced by 93% on targeted entity classes.
  • Imperative tips false positives reduced by 40% on targeted entity classes.
  • Entity recall improved by 2.5 percentage points overall.

April 2026

1 release: 0.29.0

April brought significant efficiency improvements to ingestion and restructured the search pipeline for better evolution. The focus was on reducing LLM call costs while maintaining compatibility with existing code.

Action needed
  • Episode indices are now 0-based (previously 1-based).

Highlights

  • Combined node and edge extraction now happens in a single LLM call instead of two, reducing ingestion costs.
  • Multiple episodes can be batched into one prompt for further efficiency gains.
  • Search pipeline underwent a major restructure to support upcoming improvements.
  • Episode indices are now 0-based instead of 1-based.
  • New combined extraction is opt-in, so existing ingestion code continues to work unchanged.

New

  • Combined node and edge extraction via usecombinedextraction=True parameter on extractnodesandedgesbulk.
  • Episode batching capability for multiple episodes in a single LLM prompt.
  • New tracing test coverage for the search pipeline.
  • Sagas support added.

Improved

  • Ingestion now uses fewer, smarter LLM calls overall.
  • Deduplication logic tightened for nodes and edges with looser two-entity rule in combined extraction.
  • Fact extraction improved with detail-preservation rules and broader entity capture for qualified objects.

March 2026

2 releases: 0.28.2 → 1.0.2

March brought security hardening, dependency updates, and a new hybrid LLM client option to Graphiti. The product reached version 1.0.2 with aligned core dependencies.

Highlights

  • Added GLiNER2 hybrid LLM client for improved language understanding capabilities.
  • Hardened search filters against Cypher injection attacks to improve security.
  • Updated Docker images and core dependencies to the latest versions.
  • Fixed debug logging to use edge UUIDs instead of names for clarity.

New

  • GLiNER2 hybrid LLM client support

Improved

  • Search filters now hardened against Cypher injection
  • Docker images updated to graphiti-core 0.28.1
  • Debug logs now use edge UUIDs instead of names

Fixed

  • Edge name replacement with UUID in debug logs
  • README image references corrected
  • README title and subtitle restored

February 2026

4 releases: 0.27.0 → 0.28.1

February brought significant architectural improvements to Graphiti's extraction pipeline and graph driver support. The month focused on simplifying core operations, expanding database compatibility, and addressing security concerns.

Highlights

  • Entity summaries now incorporate extracted edge facts and generate from episode data when edges don't exist.
  • Simplified extraction pipeline with batch entity summarization for improved efficiency.
  • Added support for Neptune and Kuzu graph databases alongside existing FalkorDB.
  • Replaced diskcache with sqlite-based cache to resolve a security vulnerability.
  • Refactored prompt system for better efficiency and clarity.

New

  • Batch entity summarization capability in extraction pipeline.
  • Neptune graph driver support.
  • Kuzu graph driver support.
  • Driver operations architecture allowing multiple database backends.
  • Entity summary generation from episode data when no relationships exist.

Improved

  • Extraction pipeline simplified for better maintainability.
  • Entity summaries now include extracted edge facts.
  • Prompt system refactored for efficiency and clarity.
  • FalkorDB fulltext queries now handle pipe and slash characters correctly.
  • Custom edge attributes now extracted on first episode ingestion.

Fixed

  • Duplicate edges no longer appear in node summary generation.
  • Personally identifiable information removed from log messages.
  • Security vulnerabilities in dependencies updated.
  • FalkorDB query sanitization for special characters.

January 2026

9 releases: 0.25.1 → 0.26.3

January brought several efficiency improvements and new capabilities to Graphiti, including the introduction of Sagas and better handling of large data inputs. The month focused on extraction performance and more flexible configuration options.

Highlights

  • Added Sagas for improved workflow management and orchestration.
  • Fixed entity extraction for large episode inputs with adaptive chunking to handle bigger datasets.
  • Improved edge extraction efficiency across the system.
  • Custom instructions now apply to all extraction steps for more consistent behavior.
  • Added interface override paths for direct database calls when needed.

New

  • Sagas feature for managing complex workflows.
  • Interface override paths for direct database access.
  • Adaptive chunking for entity extraction from large episodes.
  • NotImplementedError fallback for GraphOperationsInterface methods.

Improved

  • Edge extraction efficiency improvements.
  • Custom instructions now used consistently across all extraction steps.
  • Saga logic integrated into retrieveepisodes utility function.
  • DateFilter and PropertyFilter now have None defaults.

Fixed

  • Entity extraction for large episode inputs no longer fails due to size constraints.
  • Removed filterexistingduplicateofedges call from resolveextractednodes for better performance.

December 2025

3 releases: 0.24.2 → 0.25.0

December brought model updates, new filtering capabilities, and improvements to prompt customization. The month focused on stability and expanding configuration options.

Highlights

  • Property filters now available for more granular data querying
  • Custom prompt support added for personalized agent behavior
  • Model name configuration fixes ensure correct API usage
  • Triplet update capability added for knowledge graph refinement

New

  • Property filters
  • Custom prompt functionality
  • Triplet update feature

Improved

  • Model configuration handling for Gemini models
  • Edge limit handling to support more complex graphs

Fixed

  • Deprecated Gemini model replaced with current version (gemini-2.5-flash-lite)
  • Foreign language handling
  • Limited edge count issue resolved

November 2025

5 releases: 0.23.0 → 0.24.1

November brought refinements to Graphiti's core functionality and expanded platform support. The month focused on stability improvements and better integration with major AI providers.

Highlights

  • Fixed entity edge save issues to improve graph reliability
  • Integrated OpenAI structured output API for more robust response validation
  • Added dynamic token configuration for Anthropic models
  • Updated default Anthropic model to Claude Haiku 4.5

New

  • Azure OpenAI example with Neo4j for cloud deployments

Improved

  • Response validation now uses OpenAI structured output API
  • Dynamic max tokens configuration available for Anthropic models
  • Updated default Anthropic model to Claude Haiku 4.5
  • Summary character limit adjustments

Fixed

  • Entity edge save functionality

October 2025

3 releases: 0.21.0 → 1.0.0

October brought major stability improvements and the landmark 1.0.0 release, which adds multi-provider LLM and database support, a default entity ontology, and simplified deployment options.

Highlights

  • Version 1.0.0 released with support for multiple LLM providers (OpenAI, Anthropic, Gemini, Groq, Azure) and database options (FalkorDB or Neo4j).
  • New default entity ontology with 9 preconfigured entity types ready to use out of the box.
  • YAML-based configuration system with environment variable expansion and CLI overrides for easier setup.
  • All-in-one Docker image deployment option alongside support for external databases.
  • OpenTelemetry distributed tracing support added for better observability.

New

  • Multi-provider LLM support for OpenAI, Anthropic, Gemini, Groq, and Azure.
  • Multiple embedder options for flexibility in embedding generation.
  • Default entity ontology with 9 preconfigured types (Preference, Requirement, Procedure, Location, Event, Organization, Document, Topic, Object).
  • YAML configuration system with environment variable expansion and CLI overrides.
  • Health check endpoint for Docker and load balancer integration.
  • OpenTelemetry distributed tracing support.
  • OpenSearch integration for Neo4j.
  • Modular architecture using factory pattern for providers.

Improved

  • Node deduplication enhanced with deterministic matching and LLM fallbacks.
  • Entity extraction now skips processing when no fields are defined.
  • Summary generation enforced with 8 sentence limit.
  • Prompt structure refactored to move MESSAGES after instructions for clarity.
  • Prompt format changed to remove JSON indentation, reducing token usage.
  • Node extraction refactored and summary removed from attribute extraction.
  • Graph quality improvements through better edge deduplication.
  • Entity metadata now includes custom attributes and all labels in node properties.

Fixed

  • Fixed fulltext syntax error in search functionality.
  • Embedding fix to ensure correct embedding generation.
  • Duplicate edges no longer saved to the graph.
  • OpenSearch integration updates for improved compatibility.

Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. Graphiti: pricing, features and alternatives · All changelogs