Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsTo use GraphRAG on your own documents, install its Python package in an isolated environment, initialize a project, add text files and configure models, then build an index before asking questions. Choose standard or FastGraphRAG based on the graph detail you need, and choose Local, Global, Basic, or DRIFT search based on the kind of question you want answered.
What GraphRAG builds before you ask a question
GraphRAG is more than a vector-search wrapper. Its indexing pipeline processes unstructured text into structured data that can include entities, relationships, text units, community groupings, summaries, and embeddings. The resulting index supports different kinds of retrieval and analysis; the exact stages depend on the indexing method and configuration. Microsoft’s indexing overview describes the standard pipeline and its outputs.
Indexing happens before querying and can use substantial model resources. Microsoft’s Getting Started guide warns, “GraphRAG can consume a lot of LLM resources!” Start with a small dataset and inexpensive models while you learn how the pipeline behaves, rather than indexing a large corpus first. The official quickstart provides a tutorial-sized starting point.
Set up a GraphRAG project
- Create an isolated Python environment. The official quickstart calls for Python 3.10–3.12. Make a project directory and create a virtual environment there, for example with
python -m venv .venv. Activate that environment using the instructions for your operating system and shell. - Install the package. With the virtual environment active, run
pip install graphrag. - Initialize the project. Run
graphrag initfrom the project directory. Initialization creates aninputdirectory, asettings.yamlconfiguration file, and a.envfile. The quickstart walks through this sequence: Microsoft GraphRAG Getting Started. - Add source material. Place a small text file in
inputto begin. Make sure the sample documents are representative of the material and questions you ultimately care about; a toy corpus can prove that the workflow runs, but not that retrieval will work well on your real data. - Configure chat and embedding models. Initialization includes model selection. Model definitions and settings belong in the project configuration; credentials are supplied through environment-variable substitutions as appropriate for the selected configuration. The documentation does not require one provider or credential format for every setup. Keep secrets out of source files and use the configuration guidance for your installed version: YAML Configuration.
Review settings.yaml before indexing. It defines pipeline and query behavior, including model settings; the configuration reference also covers prompts, context proportions, and token limits. YAML and JSON settings are supported, but keys and defaults can change between releases. Check the configuration reference for the version you installed.
Recommended Free Tools
#1 Best Overall
- Get NVMe solid state performance with up to 1050MB/s read and 1000MB/s write speeds in a portable, high-capacity drive(1) (Based on internal testing; performance may be lower depending on host device & other factors. 1MB=1,000,000 bytes.)
- Up to 3-meter drop protection and IP65 water and dust resistance mean this tough drive can take a beating(3) (Previously rated for 2-meter drop protection and IP55 rating. Now qualified for the higher, stated specs.)
- Use the handy carabiner loop to secure it to your belt loop or backpack for extra peace of mind.
- Help keep private content private with the included password protection featuring 256‐bit AES hardware encryption.(3)
- Easily manage files and automatically free up space with the SanDisk Memory Zone app.(5). Non-Operating Temperature -20°C to 85°C
Build the index
From the project directory, run graphrag index. GraphRAG processes the input according to the selected configuration. In the standard pipeline, that can include entity and relationship extraction, optional claim extraction, community detection and report generation, and embedding generation. Parquet tables are the default output format; embeddings are written to the configured vector store. See the indexing overview for the pipeline and the architecture documentation for extension points such as input readers and vector stores.
Indexing cost is an important design constraint, not a fixed bill. Microsoft’s Methods page estimates that graph extraction accounts for “roughly 75%” of indexing cost; the page gives no publication year for that estimate. Treat it as the documentation’s approximate characterization, not a universal percentage, current price, or cost forecast for your corpus. Actual usage depends on data, models, configuration, and the chosen method. Review the indexing methods before committing to a large index.
Rank #2
- Solid state performance with up to 800MB/s read speeds in a portable drive. (Based on internal testing; performance may be lower depending on host device, interface, usage conditions and other factors. 1MB=1,000,000 bytes.)
- Back up your content and memories on a storage solution that fits seamlessly into your mobile lifestyle.
- Take it with you on your adventures—up to two-meter drop protection means this durable drive can take a beating. (Based on internal testing.)
- Secure it to your belt loop or backpack for extra peace of mind thanks to the tough rubber hook.
- From Sandisk, a brand professional photographers trust to take on assignments.
Choose an indexing method
Graph extraction determines what structure the index can support. Standard GraphRAG spends more model effort extracting and summarizing entities and relationships; FastGraphRAG trades some of that fidelity for lower indexing expense and speed.
| Method | How it builds structure | When to consider it | Trade-off |
|---|---|---|---|
| Standard GraphRAG | Uses LLM reasoning for entity and relationship extraction, their summarization, and community reports; claim extraction is optional. | When accurate entities and relationships, and a more useful graph for exploration, matter. | More indexing work and resource use than the fast alternative; the documentation characterizes extraction as a major share of indexing cost. |
| FastGraphRAG | Uses NLP noun-phrase extraction and text-unit co-occurrence links for much of the graph construction, then uses LLM generation for community reports. | When faster, less expensive indexing is the priority and noisier extraction is acceptable. | Microsoft describes the results as noisier and less directly useful for graph exploration than standard GraphRAG. |
These are qualitative trade-offs, not a published head-to-head benchmark. If your application depends on whether a named entity or relationship was extracted correctly, begin with standard indexing. If rapid, lower-cost exploration matters more, test FastGraphRAG and inspect its output against known examples. The Methods page explains the approaches.
Rank #3
- Easily store and access 2TB to content on the go with the Seagate Portable Drive, a USB external hard drive
- Designed to work with Windows or Mac computers, this external hard drive makes backup a snap just drag and drop
- To get set up, connect the portable hard drive to a computer for automatic recognition no software required
- This USB drive provides plug and play simplicity with the included 18 inch USB 3.0 cable
- The available storage capacity may vary.
Choose search based on the question
GraphRAG offers several query methods, and they answer different shapes of question. Try representative questions with the methods suited to them instead of assuming one method is best for every corpus. Microsoft’s query overview and CLI documentation describe the available options.
| Method | Best fit | How it works at a high level | Example question |
|---|---|---|---|
| Local | Questions centered on an identified entity and its context. | Combines graph-derived neighborhood information with original text chunks. | “Who is Scrooge and what are his main relationships?” |
| Global | Corpus-wide themes and synthesis. | Uses community reports in a map-reduce process to develop an answer across the corpus. | “What are the top themes in this story?” |
| Basic | Questions well served by semantic retrieval of the most relevant passages. | Provides a conventional vector-search baseline, useful for comparison with graph-oriented methods. | A focused question answerable from a small set of relevant passages. |
| DRIFT | A further supported query option to evaluate against your own questions. | Available as a query mode; consult its dedicated documentation and version-specific configuration before relying on particular behavior. | Choose a representative corpus question and compare its result with the other methods. |
Global search can use lower-level community reports to add detail, but that can increase time and LLM resource use. Microsoft’s Global Search implementation guide describes this trade-off. Do not treat a good answer from one method as evidence that all methods will perform equally well.
Rank #4
- NEARLY 2X FASTER THAN OUR PREVIOUS GENERATION(8) – move 1,000 high-res photos in under 60 seconds(6) with up to 2000MB/s transfer speeds(2).
- IP65 RATING AND UP TO 3M DROP PROTECTION(3) – protects against spills and drops.
- POCKET-SIZED – fits easily in pockets and small bags.
- SPACE TO OWN YOUR AI CONTENT – speed and capacity to download your high-res clips and photo edits.
- 256-BIT AES ENCRYPTION(4) – helps keep private files secure with password protection.
Evaluate before scaling up
Build a small test set that reflects actual use: entity-specific questions, corpus-wide questions, and questions answerable from a few passages. For each, record whether the answer is correct, whether its supporting material is relevant, and how long and resource-intensive the query feels in your environment. The documentation does not publish a universal benchmark comparing the methods, so retrieval quality has to be assessed on your corpus, prompts, models, and settings.
- Use Local for questions where identifying the right entity and its relationships is central; check that the graph context and source chunks support the answer.
- Use Global when the question asks for patterns across the collection; inspect whether the summaries preserve the details the question needs.
- Use Basic as a vector-retrieval comparison point when a question may be answered by a few semantically relevant passages.
- Include DRIFT in trials if it may suit your workflow, following the documentation for your installed version.
- Compare answer usefulness, grounding, latency, and resource use. These are evaluation dimensions, not documentation-published performance results.
Prompt tuning is part of implementation, not a cosmetic final step. Microsoft’s prompt-tuning guidance recommends adapting prompts to the corpus and task: GraphRAG Welcome and versioning guidance. Test changes with the same representative questions so that a prompt that improves one answer does not silently weaken another.
Free tools Windows power users keep installed
One-click scans. No signup required.
Best Value
- Easily store and access 5TB of content on the go with the Seagate portable drive, a USB external hard Drive
- Designed to work with Windows or Mac computers, this external hard drive makes backup a snap just drag and drop
- To get set up, connect the portable hard drive to a computer for automatic recognition software required
- This USB drive provides plug and play simplicity with the included 18 inch USB 3.0 cable
- The available storage capacity may vary.
Maintain configuration across upgrades
Keep a backup of your prompts and configuration before upgrading. The project’s versioning guidance advises running initialization between minor version bumps and using the migration notebook between major bumps; because this guidance and the package itself can change, check the current release notes and the guidance for the version you are moving to. Initialization can overwrite configuration or prompts, so do not assume your customized files will be preserved. See the project’s versioning guidance.
The architecture documentation describes extension points for readers and vector stores, but integrations can change. Verify that a specific adapter is supported by the release you use rather than relying on an older integration list: GraphRAG Architecture.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
