Release v2.0.0

TigerGraph GraphRAG v2.0.0

GraphRAG 2.0.0 is a major release built around a new agentic chat engine, structure-aware document chunking, and additive prompt customization.

What's New

• Agentic chat engine — a new default engine that plans and runs its own retrieval instead of following a fixed pipeline. It comes in two styles — Planned (lays out the whole retrieval plan up front, then answers) and Reactive (decides each step from the previous result) — chooses which methods to use, answers greetings and questions about itself directly, and cites the chunks and queries behind each answer. The Classic single-method engine remains available.
• External MCP tools — superusers can register external MCP servers (and install the Python packages they need) so the agent can call their tools during a conversation.
• Structure-aware document chunking — Markdown and HTML are split by structure, keeping each section's heading context in the chunk, rolling small sections into their parent up to the size budget, and keeping tables intact, so retrieval holds together on heading- and table-heavy documents.
• Additive prompt customization — the Customize Prompts page exposes an editable instructions-and-examples layer over fixed rules, including separate, editable retrieval strategies for the Planned and Reactive agents; edits extend behavior without dropping the rules that keep prompts working.

Changed

• Retrieval is more reliable on table-heavy and numeric content — each chunk is embedded together with a compact summary of its topic, section, and key entities, so its vector carries that context explicitly.
• Search results are bounded by relevance — hybrid and community search return the most relevant chunks up to a configurable limit, rather than everything the graph expansion reaches, reducing the context sent to the model. Tunable on the GraphRAG Configuration page.
• Answers stay grounded in document text — the agent always includes a vector search unless a question is confidently a pure structured-data request (an exact count, lookup, relationship, or aggregation).
• A streaming answer can be stopped — while the agent responds, the chat's send button becomes a stop control that ends the current response and re-enables the input.
• Chat and admin UI refinements — a clearer chat engine/style picker, bulk-clearing of older conversations, incremental chat-history loading, and the graph Compatibility Check renamed Migration Assistant.
• More reliable query installation on large graphs — graph queries install through a non-blocking request with status polling, so initialization no longer times out while queries compile.
• Ingestion survives a transient database disconnect — files whose load hits a connection error are retried once the database is reachable again, and any that still fail are named so a re-run reloads only those.

Bug Fixes

• A single oversized chunk no longer drops embeddings for the rest of its batch — over-limit embeddings are retried at progressively shorter lengths, and a vertex that still doesn't fit is skipped on its own instead of aborting the batch.
• Large ingests no longer fail on oversized upsert batches — upserts are sized to the pending work, and progress counts reflect distinct vertices and edges.
• Non-ASCII answers no longer truncate incorrectly on large context — retrieved context is measured in the same form that is sent to the model, so Japanese and other multi-byte content is sized correctly.

Upgrade notes

• Query responses default to answer-only — the query endpoints no longer return supporting sources and trace unless the caller opts in; integrations that read those fields must request them explicitly.
• Existing prompt customizations must be re-saved — with the move to additive customization, pre-existing full-prompt overrides are ignored until re-saved in the new instructions-and-examples form.
• Pick up query fixes on existing graphs — after upgrading, use Knowledge Graph Admin → Migration Assistant to re-create and reinstall shipped queries whose body has drifted, without rebuilding.

Full Changelog

v1.4.2...v2.0.0

Release v1.4.2

TigerGraph GraphRAG v1.4.2

GraphRAG 1.4.2 adds a knowledge graph compatibility check and repair tool, with reliability fixes for ingestion and sign-in.

What's New

• Knowledge graph compatibility check and repair — the Knowledge Graph Admin page can scan an existing graph for installed queries that have drifted from the shipped version or are missing, and repair them in place without rebuilding. Useful for picking up query fixes after a version upgrade on graphs that already hold data.

Changed

• Shipped query fixes apply automatically — on initialization, queries whose installed body has drifted from the shipped version are re-created, so improvements take effect after an upgrade without a manual reinstall.
• Clearer upload failures — when some files fail to upload, the document ingestion dialog names the affected files and the reason instead of failing the batch opaquely.

Bug Fixes

• Documents with spaces or mixed case in filenames now ingest reliably — they are stored under one consistent id across every ingest path instead of being silently skipped or duplicated, so they become retrievable and participate in entity extraction.
• Interrupted rebuilds recover cleanly — chunks left unfinished by a crashed or cancelled run are reconciled on the next run, eliminating chunks left without content and the spurious "missing content" warnings they caused.
• Sign-in respects configured TigerGraph ports — logging in works on deployments where TigerGraph runs on non-default GS/RESTPP ports.

Full Changelog

v1.4.1...v1.4.2

Release v1.4.1

TigerGraph GraphRAG v1.4.1

GraphRAG 1.4.1 adds token-based sign-in and a pre-flight upload conflict check, plus a wave of ingestion-UX, image-description, and reliability fixes.

What's New

• Token-based sign-in — sign in with an API token or shared secret in addition to username and password. The signed-in identity, including real username and roles, is shown consistently across the UI.
• Pre-flight upload conflict check — uploads warn about filename collisions before sending the bytes and offer a one-click choice to replace or skip the conflicting files.

Changed

• Per-graph image-description controls — new extract_images toggle and min_image_dim_px threshold on the GraphRAG Configuration page, settable globally or per graph. Disable image description on graphs that don't need it, or raise the dimension threshold to skip more small images.
• Faster image-heavy PDF ingestion — PDFs with many images ingest significantly faster, decorative images are excluded from the graph, and image descriptions are written in the document's own language.
• Multilingual answer fidelity — the chatbot answers in the question's language, quotes exact numbers and units from the source, and lists candidates before stating a maximum, minimum, or other comparison conclusion.
• Community search falls back to hybrid on miss or failure — auto-selected community questions that find no relevant summaries are retried with hybrid search instead of returning a "couldn't find" answer.

Bug Fixes

• Sample documents disappeared from the upload dialog after schema extraction — they now stay visible in the dialog after extraction. Re-uploading the same filename triggers a clean re-extract on the next ingest.
• Chat disconnected on vector-store errors — chat stays open through transient vector-store problems. Graph-traversal questions answer normally; only questions that strictly need vector search fail per-question with a clear message.
• Ingest button enabled when reopening the dialog mid-conversion — could trigger a double upload. The button now correctly reflects the current job state on reopen, and the uploaded-files list refreshes.

Full Changelog

v1.4.0...v1.4.1

Release v1.4.0

TigerGraph GraphRAG v1.4.0

GraphRAG 1.4.0 introduces schema-aware knowledge graphs, an auto retrieval method selector, and a Trace Logs UI.

What's New

• Schema-aware knowledge graphs — declare a domain schema during graph creation (skip / generate from sample documents / paste GSQL, with a per-type / per-attribute editor before apply). Ingestion then populates your declared vertex and edge types directly.
• Auto retrieval method selection — new "Auto" option in the chat dropdown picks among Similarity / Contextual / Hybrid / Community per question. The choice is shown as an inline chip below each reply.
• Trace Logs UI — admin view showing per-node inputs / outputs, durations, citations, and token usage for every chat reply. Opens inline as a modal from the chat.
• Per-stage progress for graph rebuild — the Rebuild dialog shows each phase as it runs.

Changed

• Default chat retrieval method is now auto instead of hybridsearch.
• Prompts are now built into the container. No more shipping a prompt-file directory inside the image; only prompts you customize through the Customize Prompts page are written to disk.
• Sessions no longer expire while long operations are running (init, schema extraction, upload, conversion, ingest, rebuild).
• Navigation locks during chat streaming — Setup, conversation history, "New Chat", and the chat input are disabled while an answer is being generated. Logout stays enabled.

Bug Fixes

• Images in chat render correctly — multi-line image captions no longer break the markdown.
• Fewer spurious rebuild failures — rebuilds no longer report false failures when TigerGraph completed cleanly, and survive transient empty-chunk reads.
• LLM-generated schemas apply cleanly even when proposed attribute names collide with GSQL reserved words.
• Ingest no longer fails with Data path not found: None when the Ingest Documents button is clicked before the two-step flow.
• Chat shows a clear error instead of disconnecting silently when the embedding store is unavailable.
• Faster container restarts when an embedding store is already installed.

Full Changelog

v1.3.1...v1.4.0

Release v1.3.1

What's New

• PNG/GIF Image Support — Text extractors now handle .png and .gif files for multimodal ingestion.
• Loading Job Auto-Recreation — Missing loading jobs are automatically recreated before ingestion.

Changed

• Upgraded pyTigerGraph to >=2.0.3.
• Deferred community query installation to community detection phase to reduce ECC init timeout.

Bug Fixes

• Entity description unbounded growth — fuzzy dedup (SequenceMatcher, threshold 0.85) prevents near-duplicates accumulating in SET<STRING>.
• WebSocket crash on early disconnect — WebSocketDisconnect caught during auth phase.
• stream_community response size — projects only vertex ID to stay within 5MB limit.

Full Changelog: v1.3.0...v1.3.1

Release v1.3.0

TigerGraph GraphRAG v1.3.0

What's New

• Admin Configuration UI
  • Role-based configuration pages for DB, LLM providers, and GraphRAG settings with secret masking and graph admin access control.
• Unified LLM Config Save/Test
  • Save and test connection share the same config resolution path. Top-level parameters (auth, region_name) are normalized and per-service duplicates are stripped automatically.
• Per-Graph Config Overrides
  • Graph-specific LLM and GraphRAG settings store only the delta against global config. Superadmins can override all services; graph admins can override the chatbot model.
• Per-Graph Chatbot LLM Override
  • Configure a distinct chat_service per graph with automatic inheritance from completion_service.
• Multimodal Inherit Option
  • "Use same model as completion service" checkbox in LLM config UI for both single and multi-provider modes.
• Centralized LLM Token Usage Tracking
  • All LLM call sites migrated to invoke_with_parser / ainvoke_with_parser with structured and plain text response support.
• Retriever Scoring for All Retriever Types
  • Similarity, Hybrid, and Sibling retrievers now score and rank context chunks when combine=False, matching CommunityRetriever behavior.
• User-Customized Prompts
  • Per-graph prompt overrides persisted under configs/ across container restarts.
• Session Idle Timeout
  • 1-hour idle timeout with auto-clear; timer pauses during long-running operations.
• Auth Guard on All UI Routes
  • RequireAuth wrapper redirects unauthenticated users to login.
• Greeting Detection
  • Agent router responds directly to greetings, farewells, and thanks without invoking search.
• GraphRAG Config UI Fields
  • Search parameters (top_k, num_hops, community_level, doc_only) and advanced ingestion settings exposed in the UI.

Changed

• Config resolution uses get_xxx_config(graphname) getters everywhere instead of direct llm_config access.
• Atomic config file writes with file locking to prevent race conditions.
• Chat history messages display instantly without typewriter animation.
• default_concurrency replaces tg_concurrency in graphrag_config.

Bug Fixes

• Bedrock multimodal connection test — replaced 1x1 test PNG with 20x20 PNG accepted by Bedrock validation.
• Provider-aware image format — GenAI/VertexAI use image_url; Bedrock/Anthropic use type:"image" with source block.
• Null service values stripped before config reload (null = inherit, key should be absent).
• Login error messages — proper feedback based on HTTP status codes.
• Bedrock model names no longer trigger token calculator warnings.
• Config reload no longer clears in-memory state during concurrent requests.
• JSON parsing fallback — handles LLM responses wrapped in preamble text or markdown code fences.
• Cypher/GSQL output validation before query execution to prevent garbage queries.

Full Changelog: v1.2.0...v1.3.0

v1.2.0

TigerGraph GraphRAG v1.2.0

What's New

• Knowledge Graph Management UI
  • Graph initialization, document ingestion, and knowledge graph rebuild with status tracking, confirm dialogs, and rebuild lock management.
• Server-Side & Batch Ingestion
  • Local folder processing, S3 data ingest, BDA ingestion, batch upload, and re-ingestion of previously processed files without reprocessing.
• Image Processing
  • Extract and display images from PDFs using multi-modal LLMs, with logo identification and in-answer image rendering.
• Chat History
  • Conversation history displayed in the UI and used as context for follow-up questions.
• Append Schema to Existing Graph
  • GraphRAG schema can now be added to an existing TigerGraph graph.
• Separate Chat Model
  • Configure a distinct chat_model for the chatbot, separate from the extraction LLM.
• Azure OpenAI Improvements
  • Embedding dimension support and other Azure service upgrades.

Schema Changes

• Removed Entity Embedding
  • Only DocumentChunk and Community retain the embedding vector attribute. Removed from Entity and other vertex types to reduce storage overhead.

Bug Fixes

• Non-ASCII / CJK Text Support
  • Fixed text corruption for non-ASCII characters (Chinese, Japanese, Korean, etc.) during chunking and extraction. Added CJK sentence-boundary separators for better chunking quality.
• Rebuild Lock & Status
  • Fixed race conditions, stale lock messages, and UI stuck on "Checking rebuild status...". Prevented ingestion during active rebuilds.
• Memory-Efficient Data Loading
  • Switched to runLoadingJobWithFile to reduce memory consumption during ingestion.
• Token Truncation
  • Revised token calculation to prevent context overflow.

Full Changelog: v1.1.0...v1.2.0

v1.1.0

TigerGraph GraphRAG v1.1.0

What's New

• AWS Bedrock Integration
  • Full support for AWS Bedrock as LLM provider, including Bedrock Data Automation (BDA) for document processing and S3-based data ingestion pipeline with presigned URL support.
• Gemini & Ollama Support
  • Added Google Gemini as a supported LLM provider and Ollama as a supported embedding service.
• Kubernetes Deployment
  • Added K8S deployment files for production deployment.
• Enhanced Community Search
  • Improved community search retriever for better answer quality.

Improvements

• TigerGraph v4.2.0/v4.2.1 Compatibility
  • Ensured compatibility with both TigerGraph v4.2.0 and v4.2.1, including loading job fixes.
• LLM & Embedding Enhancements
  • Improved OpenCypher query generation for faster, more accurate responses. Upgraded GenAI library with embedding fixes. Added LLM token usage logging.
• Chunking & Retrieval
  • Added chunk size control for the markdown chunker. Revised vector search retrievers for better retrieval quality.

Bug Fixes

• Async Mode Authentication
  • Fixed errors in async mode with bearer token authentication.

Full Changelog: v1.0.0...v1.1.0