Knowledge Graph / GraphRAG
Tool | Category | Segment | Platform / Tool | Plan / License | Monthly Price USD | Pricing Model | Free Tier / OSS | Included Usage / Limits | Graph Construction / Knowledge Model | Retrieval / Reasoning | GraphRAG / Context Features | Integrations / Frameworks | Deployment / Hosting | Security / Privacy | Team / Governance | Best Fit | Main Limits / Caveats |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
No tagline | Knowledge Graph / GraphRAG | Property-graph retrieval layer | LlamaIndex Property Graph Index | MIT | $0 software | Open-source framework component; model, embedding and graph-store costs separate | ✓ | No software usage cap | PropertyGraphIndex extracts graph nodes and relations into a simple in-memory store or external property-graph backends | Composable PGRetriever supports LLM synonyms, vector context, TextToCypher, Cypher templates and custom sub-retrievers | Graph construction and retrieval can be combined with the broader LlamaIndex ingestion, query-engine and agent ecosystem | Python, LlamaIndex, Neo4j and other property-graph stores plus configurable LLM/embedding providers | Embedded/local or customer-managed infrastructure | Security and data handling depend on the selected graph store and model providers | No independent hosted governance in the OSS index layer | Existing LlamaIndex applications that need property-graph extraction and modular retrievers | This is a component of a broader framework, not a turnkey GraphRAG product; quality depends on extraction prompts and graph schema |
No tagline | Knowledge Graph / GraphRAG | Lightweight GraphRAG framework | LightRAG | MIT | $0 software | Open-source server/core; model, embedding, reranker and storage costs separate | ✓ | No software usage cap; REST server, Web UI and local core are included | Builds an entity-relation knowledge graph linked to chunks, vectors, document status and LLM response caches | Local, global, hybrid, mix and naive query modes; optional reranker and graph-plus-vector retrieval | Incremental ingestion, entity and relation management, deletion/merge, citations, graph export, evaluation and observability hooks | OpenAI, Azure, Gemini, Ollama, Hugging Face and LlamaIndex providers; NetworkX, Neo4j, PostgreSQL AGE, Memgraph, MongoDB and Redis storage options | Self-hosted Python, REST server, Docker or Kubernetes | Can run fully local; external model and storage services determine the final data path | No hosted RBAC in OSS; operational controls depend on the chosen deployment stack | Teams wanting a practical self-hosted GraphRAG server with multiple storage and model backends | Requires coordinated KV, vector, graph and document-status storage; embedding model changes can require rebuilding data |
No tagline | Knowledge Graph / GraphRAG | Managed temporal context graph | Zep | Commercial cloud SaaS | $0 | Credit-based cloud pricing; paid Flex starts at $125/month | ✓ | 1,000 credits/month, 2 projects, 5 custom entity/edge types and variable rate limits; free credits do not roll over | Managed temporal context graphs ingest messages, JSON and text into evolving entities, facts and relationships | Low-latency relationship-aware retrieval with incremental updates and temporal fact validity | Managed users, threads, context assembly, API logs, graph visualization and production SDKs | Python, TypeScript and Go SDKs plus common agent frameworks | Managed cloud; Enterprise supports BYOK, BYOM and BYOC in AWS | SOC 2 Type II; HIPAA BAA, audit logs and stronger residency controls are Enterprise features | Free includes 2 projects; Flex includes 5 projects; Enterprise adds SLA, support and governance | Production agents that need managed real-time GraphRAG and temporal memory | Credits are based on episode size; free rate limits are variable and Graphiti flexibility is reduced in exchange for managed behavior |
No tagline | Knowledge Graph / GraphRAG | GraphRAG framework | Microsoft GraphRAG | MIT | $0 software | Open-source framework; LLM, embedding, storage and compute costs separate | ✓ | No software usage cap; current repository warns that indexing can be expensive | LLM extraction of entities, relationships and claims, followed by community detection and community reports over a document corpus | Local, global and graph-aware search modes assemble evidence from entities, relationships, communities and source text | CLI and Python pipeline, prompt tuning, structured artifacts and a unified search application | Python, OpenAI/Azure-oriented model configuration, Parquet artifacts and custom data pipelines | Local or customer-managed infrastructure | Data path is controlled by the selected model and storage providers; no Microsoft-hosted service is included | Source-control governance only; repository states the code is a demonstration and not an officially supported Microsoft offering | Corpus-wide thematic discovery and multi-hop questions over large private text collections | Indexing cost can be high; configuration changes across versions may require migration or re-indexing |
No tagline | Knowledge Graph / GraphRAG | Minimal GraphRAG implementation | nano-graphrag | MIT | $0 software | Open-source Python library; model and storage costs separate | ✓ | No software usage cap; package is roughly 1,100 lines excluding tests and prompts | Extracts graph entities and communities from text with compact local persistence and optional Neo4j/FAISS components | Local and global GraphRAG search plus optional naive RAG; async APIs and incremental insert are supported | Small typed codebase intended for inspection, modification and experimentation | Python, OpenAI/Azure, Ollama, FAISS and Neo4j-oriented extensions | Local/customer infrastructure | Can stay local with local models and stores; external APIs change the data path | No team control plane or managed governance | Researchers and developers who need a compact, hackable GraphRAG reference implementation | Does not implement every Microsoft GraphRAG feature; community reports are recomputed after inserts and production hardening is limited |
No tagline | Knowledge Graph / GraphRAG | Multimodal GraphRAG framework | RAG-Anything | MIT | $0 software | Open-source framework; parsing, vision, LLM, embedding and storage costs separate | ✓ | No software usage cap | Builds a multimodal knowledge graph from text, images, tables, equations and document hierarchy | Fuses vector similarity with graph traversal and modality-aware ranking | End-to-end document parsing, cross-modal relationships, hierarchy preservation and multimodal query answering | Python, LightRAG, MinerU and configurable vision/LLM/embedding providers | Self-hosted application or service | Local deployment is possible, but OCR, vision and model services may receive document content | No built-in enterprise governance in the OSS project | Complex PDF, Office and multimodal knowledge bases where relationships cross text, images and tables | More dependencies and processing stages than text-only GraphRAG; parser and model quality strongly affect the graph |
No tagline | Knowledge Graph / GraphRAG | Semantic context graph platform | TrustGraph | Apache-2.0 | $0 software | Open-source full stack; model, storage and infrastructure costs separate | ✓ | No software usage cap | Builds context graphs using RDF/OWL-oriented semantics, ontologies, schemas and governed document ingestion | GraphRAG queries expose provenance and explainability; graph neighborhoods and semantic workflows support agent context | Agent console, GraphRAG view, 3D context explorer, ontology workbench, schema tools, flows and prompt editing | Python services, TypeScript UI clients, SPARQL/RDF/OWL/SHACL ecosystem and configurable model infrastructure | Containerized deployment through Docker, Podman, Minikube or Kubernetes | Designed for private sovereign deployments with customer control of models and data | Workspace management and workbench tools are included; identity/RBAC depth depends on the deployed configuration | Organizations needing an explainable, ontology-heavy GraphRAG platform rather than a small library | Large multi-service platform with more operational complexity than embedded retrieval SDKs |
No tagline | Knowledge Graph / GraphRAG | Graph memory platform | Cognee | Apache-2.0 OSS; commercial cloud | $35/month cloud; $0 OSS | Open-source self-hosting or fixed managed plans with data/document top-ups | ✓ | Developer includes 1 user, 1,000 documents or 1 GB and 10,000 API calls; OSS has no software cap; 14-day cloud trial is documented | Cognify pipelines combine embeddings, entities, relationships, ontologies, provenance and feedback into graph memory | Recall auto-routes across session memory, semantic search and graph-connected knowledge | Remember/recall/forget/improve API, multimodal ingestion, evaluations, MCP, UI and agent memory plugins | Python, MCP, Claude Code, LangGraph, OpenClaw, Kuzu, LanceDB, PostgreSQL and many data sources | Local/self-hosted, Cognee Cloud, Modal, Railway, Fly.io, Render or enterprise on-prem | Self-hosting supports air-gapped control; cloud provides permissions, RBAC and audit-oriented features | Developer is single-user; Team is $200/month for 10 users and multi-tenant memory grouping | Teams needing a broad agent-memory layer that blends knowledge graphs and vector retrieval | Broader than GraphRAG alone; cloud document/API allowances and top-up pricing must be modeled for production volume |
No tagline | Knowledge Graph / GraphRAG | Graph retrieval layer | DataStax Graph RAG | Apache-2.0 | $0 software | Open-source retriever library; vector store and model costs separate | ✓ | No software usage cap | Does not build a separate graph database; graph edges are defined through metadata relationships already stored with vectors | Combines vector similarity with traversal of metadata properties to retrieve connected chunks | Generic graph-retriever core and LangChain-specific GraphRetriever wrapper | Python, LangChain and supported vector-store adapters including Astra DB patterns | Embedded in customer applications | Data remains in the configured vector store and application/model stack | No hosted governance or control plane | Adding relationship traversal to an existing vector-store RAG system without adopting a graph database | Relationship metadata must already exist; backend support is limited and the last public release listed is April 2025 |
No tagline | Knowledge Graph / GraphRAG | Local GraphRAG application | GraphRAG Local UI | MIT | $0 software | Open-source local application; local hardware and optional model API costs separate | ✓ | No software usage cap | Adapts Microsoft GraphRAG indexing and prompt tuning for local models and local artifacts | Supports global, local and direct chat queries through a FastAPI backend | Gradio indexing UI, prompt tuning, file management, output exploration and 2D/3D graph visualization | Python, FastAPI, Gradio, Ollama and OpenAI-compatible local endpoints | Local/self-hosted; primarily documented for workstation use | Can keep documents and models local; optional external endpoints change the data path | No multi-user governance; collaborative features remain roadmap items | Developers wanting a visual local GraphRAG workbench around Microsoft GraphRAG concepts | Repository says updates are slow, the UI architecture is transitioning and testing has focused mainly on a Mac Studio M2 |
No tagline | Knowledge Graph / GraphRAG | Knowledge-graph reasoning research | Think-on-Graph | Apache-2.0 stated in README | $0 software | Open-source research implementation; KG endpoints and model calls separate | ✓ | No software usage cap | Operates on existing large knowledge graphs such as Freebase or Wikidata rather than constructing a document graph | Iteratively explores candidate relations and entities with LLM-guided pruning to form reasoning paths | Responsible, interpretable graph traversal for multi-hop knowledge-graph question answering | Python, Freebase, Wikidata and configured LLM APIs | Local research environment plus external or local KG service | Queries and intermediate paths follow the configured KG and LLM services | No production team controls or hosted service | Researchers comparing explicit KG traversal and multi-hop reasoning methods | Research code and datasets require setup; not a document-ingestion GraphRAG framework or supported production SDK |
No tagline | Knowledge Graph / GraphRAG | Biomedical KG-RAG framework | KG-RAG | Apache-2.0 | $0 software | Open-source biomedical framework; SPOKE access, model and infrastructure costs separate | ✓ | No software usage cap | Uses the SPOKE biomedical knowledge graph and disease/entity recognition to ground queries in curated relationships | Retrieves graph neighborhoods and semantically prunes context before LLM generation | Evidence-focused biomedical answers with provenance, configurable graph depth and GPT or local Llama workflows | Python, Neo4j/SPOKE, sentence transformers, OpenAI and local Llama examples | Local/customer infrastructure; a specialized API is also documented by the project team | Can self-host the pipeline; biomedical source licenses and model-provider policies still apply | No general team governance; specialized institutional deployments manage access separately | Biomedical question answering that can rely on SPOKE concepts and relationships | Domain-specific rather than general purpose; some releases and examples are older and require external biomedical graph access |
No tagline | Knowledge Graph / GraphRAG | GraphRAG SDK | FalkorDB GraphRAG SDK | Apache-2.0 | $0 software | Open-source SDK; FalkorDB, model, embedding and hosting costs separate | ✓ | No software usage cap | Ingests raw documents into configurable entity/relation schemas with entity resolution, deduplication, provenance and incremental updates | Combines vector search, full-text search, Cypher, relationship expansion, optional Text2Cypher and local cosine reranking | Cited answers, per-tenant graph names, change application APIs, crash-safe updates and pluggable strategies/providers | Python, FalkorDB, LiteLLM and custom LLM/embedder providers | Self-hosted SDK with local or managed FalkorDB | Data control follows the FalkorDB and model deployment; source provenance is retained through MENTIONS edges | Graph-name isolation supports multi-tenant patterns; broader RBAC depends on the database/deployment | Production-oriented GraphRAG pipelines that want a focused SDK and FalkorDB performance | Requires FalkorDB and a finalize step whose cross-document deduplication cost grows with graph size |
No tagline | Knowledge Graph / GraphRAG | Graph memory and multi-hop retrieval | HippoRAG | MIT | $0 software | Open-source research package; LLM, embedding and GPU costs separate | ✓ | No software usage cap | Uses OpenIE outputs to build a graph-like long-term memory connecting passages, entities and facts | Personalized PageRank and graph-based association support multi-hop retrieval, sense-making and continual knowledge integration | Separate retrieval and QA APIs, incremental updates, deletion, reranking and evaluation workflows | Python, OpenAI/Azure/Bedrock-compatible paths, local vLLM and selected embedding models | Local research or customer infrastructure | Local deployment is supported; hosted model providers receive configured prompts and documents | No hosted team governance or production control plane | Multi-hop QA and agent memory research where associative retrieval is more important than corpus summarization | Research-oriented setup can require GPUs and specific embedding/OpenIE configurations; production operations are user-owned |
No tagline | Knowledge Graph / GraphRAG | Adaptive GraphRAG framework | Fast GraphRAG | MIT | $0 software; managed service has 100 free requests/month | Open-source library or usage-based managed service; public post-free pricing is not itemized in the repository | ✓ | No OSS software cap; first 100 managed requests are free each month | Automatically generates and refines a domain graph and supports incremental updates as source data changes | Personalized PageRank-based exploration targets relevant graph regions for interpretable retrieval | Promptable graph behavior, async typed APIs, dynamic data updates and lower-cost indexing goals | Python, PyPI, OpenAI-compatible models and configurable local infrastructure | Self-hosted library or Circlemind managed service | Self-hosting controls the data path; managed-service privacy and residency require a service review | OSS has no team governance; managed controls are not publicly detailed | Applications needing faster incremental graph retrieval without the full Microsoft GraphRAG pipeline | Managed pricing after the free allowance is not public; framework maturity and backend choices still need production validation |
No tagline | Knowledge Graph / GraphRAG | Knowledge-augmented reasoning framework | KAG | Apache-2.0 | $0 software | Open-source framework built on OpenSPG; model and infrastructure costs separate | ✓ | No software usage cap | Combines schema-free extraction with schema-constrained professional knowledge, semantic alignment and graph-to-chunk mutual indexing | Logical-form-guided hybrid reasoning and retrieval supports factual, logical and multi-hop questions | Domain ontology modeling, knowledge construction, reasoning operators and professional knowledge-base workflows | Python, OpenSPG engine, graph/search components and configurable LLM providers | Self-hosted/customer-managed | Private deployment is possible; model and search backends determine external data exposure | Governance follows the OpenSPG deployment; no standalone hosted team plan is included | Professional-domain knowledge bases where rules, schemas and multi-hop factual reasoning matter | Operationally heavier than lightweight GraphRAG libraries and depends on the OpenSPG ecosystem |
No tagline | Knowledge Graph / GraphRAG | Small-model GraphRAG framework | MiniRAG | MIT | $0 software | Open-source research framework; model, graph store and compute costs separate | ✓ | No software usage cap; repository reports support for API/Docker and 10+ heterogeneous graph databases | Creates a heterogeneous graph linking text chunks and named entities with a compact index | Lightweight topology-enhanced retrieval is designed to work with small and open-source language models | On-device-oriented design, included benchmark data and graph visualization components | Python, LightRAG-derived components, Neo4j, PostgreSQL, TiDB and other graph backends | Local, API or Docker deployment | Can use local small models and self-hosted graph storage | No hosted collaboration or policy layer | Resource-constrained and on-device experiments requiring graph retrieval with small models | Early project with a small number of releases; production scalability and backend compatibility need validation |
No tagline | Knowledge Graph / GraphRAG | Temporal context graph framework | Graphiti | Apache-2.0 | $0 software | Open-source framework; graph database, model and infrastructure costs separate | ✓ | No software usage cap | Builds temporal context graphs with entities, facts, validity windows, episodes, provenance and custom Pydantic ontologies | Hybrid semantic, keyword and graph-traversal search with graph-distance reranking and historical queries | Continuous incremental updates, contradiction handling, temporal history, provenance and MCP/REST options | Python, Neo4j, FalkorDB, Kuzu, Amazon Neptune, OpenAI, Anthropic, Groq, Gemini, Ollama and LangGraph examples | Self-hosted only for Graphiti; managed deployment is the separate Zep product | Repository says anonymous environment/configuration telemetry is enabled by default but can be disabled; graph content and queries are not collected | No built-in users, threads, dashboard or enterprise governance in the OSS engine | Dynamic agent context and memory where facts change over time | Requires an external graph backend and surrounding production tooling; default ingestion can trigger model-provider rate limits |
No tagline | Knowledge Graph / GraphRAG | Graph retrieval SDK | Neo4j GraphRAG for Python | Apache-2.0 | $0 software | Open-source Python package; Neo4j hosting, model and embedding costs separate | ✓ | No package usage cap | Knowledge Graph Builder pipelines can extract entities and relationships into Neo4j; existing graphs can be used directly | Vector, hybrid, graph traversal, Text2Cypher and external-vector-store retrievers feed a common GraphRAG generation pipeline | First-party Neo4j package with retrievers, LLM/embedder adapters, KG construction and source-aware generation | Python, Neo4j, OpenAI, Azure, Gemini, Cohere, Anthropic, Mistral, Bedrock, Ollama, Weaviate, Pinecone and Qdrant | Self-hosted Neo4j or Neo4j Aura plus application infrastructure | Security, encryption, tenancy and residency follow the selected Neo4j deployment and model providers | No separate SaaS governance in the library; Aura and Neo4j Enterprise controls apply to the database layer | Python teams standardizing GraphRAG on Neo4j with supported retriever patterns | Tied to Neo4j for graph operations; some KG Builder components are experimental and ANN vector search is approximate |
No tagline | Knowledge Graph / GraphRAG | Medical GraphRAG research | Medical Graph RAG | MIT | $0 software | Open-source research system; model, API, dataset and infrastructure costs separate | ✓ | No software usage cap | Constructs hierarchical medical graphs linking private patient data, literature/books and dictionary/ontology sources such as UMLS | Retrieves evidence across graph levels for grounded medical question answering | Agentic chunking, graph construction, hierarchical linking, PubMed-oriented demo and Docker example | Python, Docker, OpenAI, NCBI/PubMed, UMLS and included GraphRAG components | Local research deployment or Docker demo | Medical data requires strict local governance; external OpenAI/NCBI calls can expose query or document content | No clinical access-control or compliance layer is provided by the research code | Research on evidence-backed medical retrieval across private and public knowledge layers | Not a clinical product; source datasets can be licensed/restricted and outputs require medical validation |
No tagline | Knowledge Graph / GraphRAG | GNN graph retriever research | GNN-RAG | No explicit license detected | $0 code; commercial rights unclear | Public research repository; training, KG and inference costs separate | Code available | No software usage cap | Uses a graph neural network retriever over an existing knowledge graph to identify relevant entities, relations and paths | Retrieved graph evidence is supplied to an LLM for knowledge-graph reasoning and QA | Separates graph retrieval from language generation and includes research evaluation workflows | Python, PyTorch/graph ML tooling, KGQA datasets and LLM components | Local research environment | Data remains in the configured research stack unless external model APIs are used | No team governance or hosted service | Graph ML researchers testing learned subgraph/path retrieval for LLM reasoning | No explicit repository license was detected, so reuse beyond research needs legal review; setup is benchmark-oriented |
No tagline | Knowledge Graph / GraphRAG | Graph foundation model retriever | GFM-RAG | Apache-2.0 | $0 software | Open-source package and pretrained retrievers; compute, model and graph-building costs separate | ✓ | No software usage cap | Builds a graph index from documents or loads a user-provided nodes/relations/edges graph format | A pretrained graph foundation model retrieves relevant documents by reasoning over graph structure and can be fine-tuned on query-document pairs | Bring-your-own-graph path, reusable G-reasoner models, retrieval API and QA/evaluation workflows | Python, PyPI, Hydra, PyTorch, Hugging Face models and configurable graph constructors | Local research or customer GPU infrastructure | Can operate locally; model downloads and any external LLM generation follow provider policies | No hosted team governance | Teams researching learned graph retrievers that transfer across unseen datasets | Heavier ML stack than rule-based graph retrieval; GPU and model compatibility need validation for production |
No tagline | Knowledge Graph / GraphRAG | Markdown graph retrieval layer | IWE | Apache-2.0 | $0 software | Open-source local CLI/LSP/MCP tool; no model or database service is bundled | ✓ | No software usage cap; repository reports processing 20,000 files in under a second | Turns Markdown files and links into a local knowledge graph without a vector database | CLI and MCP tools let agents search, read, navigate and refactor linked notes | Git-versioned plain-text knowledge, editor navigation, structured agent access and no built-in AI dependency | Rust CLI, MCP, VS Code, Neovim, Zed, Helix, Claude, Codex and Gemini | Local-only/customer filesystem | No cloud, database or vendor lock-in; files remain under user control | Governance follows filesystem permissions and Git review workflows | Developer and research knowledge bases that are already organized as linked Markdown | Not an automatic document-to-KG or answer-generation system; graph quality depends on authored links and note structure |
No tagline | Knowledge Graph / GraphRAG | Ontology-grounded hypergraph RAG | OG-RAG | MIT | $0 software | Open-source research implementation; model and infrastructure costs separate | ✓ | No software usage cap | Maps domain documents and expert ontologies into hypergraph representations of related factual knowledge | Optimization-based retrieval selects a compact set of hyperedges that covers query-relevant facts | Ontology grounding, fact-focused context assembly and traceable attribution for specialized workflows | Python, domain ontologies, embeddings and configured LLM services | Local/customer research infrastructure | Can be self-hosted; model-provider and ontology-source policies determine data handling | No hosted governance or team workflow | High-stakes domain workflows where ontology-defined facts and attribution matter | Requires a usable domain ontology and mapping process; repository has no packaged releases and remains research-oriented |