Humata
ProductivitySearch, summarize and compare uploaded documents with source-linked answers. Humata adds team-oriented document access and page-based plans for ongoing knowledge work.
Questions across filesCited passages
STANFORD STORM / RESEARCH BEFORE THE DRAFT
Stanford STORM
An open-source research-and-writing system that gathers sources, develops an outline and drafts cited articles. It is useful for pre-writing and topic exploration, with model and search costs separate from the free software.
Official STORM demo interface showing an article beside its outline and references; an upstream project exampleExplore the image ↓
What you can do with it
What this tool is for
STORM is a research system for producing Wikipedia-style article drafts from a topic. It separates source gathering and outline development from the writing stage, making it relevant when you need to understand what a topic contains before deciding how to explain it.
The project explicitly says its output is not publication-ready. That is a useful expectation: the generated draft is a map of questions and material to inspect, not a finished authority. A strong workflow keeps the sources visible, revises the outline and checks whether each important claim is actually supported.
01
An unfamiliar topic briefing
02
An editorial pre-writing packet
03
A research-method demonstration
04
An internal knowledge prototype
A closer look at the product
Official product image. Click to inspect the details.
Product interface and examples
Key features, practical tasks
Capabilities in practice
- Research before writing: STORM gathers references before generating the article. Use this stage to see whether the topic has been framed too broadly or whether the available sources cover only one part of the question.
- Perspective-guided questions: The method explores different perspectives and simulates a writer–expert conversation. This can suggest questions you had not considered, but it does not remove the need to judge whether those perspectives are relevant or balanced.
- Outline generation: An outline organizes the collected material before the full draft. Review the order and scope of sections here; fixing a weak structure early is easier than polishing many paragraphs that do not serve the intended reader.
- Citation-bearing drafts: The writing stage uses references and an outline to generate an article with citations. Inspect the cited passages for the claims you retain, because a citation marker alone does not establish accurate support.
- Co-STORM collaboration: The related Co-STORM workflow allows human participation in an agent discussion and maintains a concept map. It is useful for steering exploration, while the participant remains responsible for deciding which questions matter.
- Custom models and retrieval: The project provides modular model and search components and examples using open-weight models. Choose a configuration according to quality, cost and data requirements rather than assuming every example uses the same backend.
- Private-corpus experiments: An example supports a supplied corpus through a vector retriever, and a minimal interface displays articles and references together. A private corpus still needs appropriate access controls and carefully configured model processing.
Get started, step by step
- 01
Define the question and reader
- 02
Configure models and retrieval
- 03
Review evidence and outline
- 04
Edit and verify the final draft
A practical first workflow
A practical first workflow
- Define the question and reader. Choose a topic narrow enough to review, state the audience and decide what should be outside the scope. Avoid asking for an exhaustive report when the sources or budget are limited.
- Configure models and retrieval. Follow the official setup with a deliberate model and search or corpus choice. Keep credentials protected and estimate the external usage that a multi-stage run can create.
- Review evidence and outline. Look at the research results before accepting the structure. Remove irrelevant sections, identify missing viewpoints and mark claims that need a stronger source.
- Edit and verify the final draft. Read important citations, correct unsupported statements and write a clear explanation for the intended reader. Publish only after the article has undergone the necessary human editorial review.
Example prompt or task: Research how small teams can maintain an internal technical knowledge base. Cover information ownership, update routines and retrieval limits; separate documented facts from recommendations and preserve the source for each key claim.
This is an example topic brief. The configured STORM pipeline determines how the topic and additional guidance are supplied.
Find the right plan for you
Free self-hosted software
Explore official plans ↗Free software Model and retrieval costs are separate
A public research preview is not a guarantee of unlimited hosted capacity. Self-hosting still needs configured model and retrieval services.
What is free and what can cost extra
The project is MIT-licensed software, so downloading and self-hosting the code does not require a software subscription. Running it still needs language models and retrieval; hosted model calls, search APIs, embeddings, vector storage or infrastructure can incur charges.
The repository includes open-weight and custom-corpus examples, but a local configuration consumes hardware and requires setup. Public research-preview availability is separate from operating your own instance. Set a practical budget before using many perspectives or repeated full-length article runs.
A few things worth knowing
Frequently asked questions
Is STORM free software?
Yes. The project uses the MIT license.
Does free software include free model calls?
No. Inference, retrieval and hosting are separate.
Are its articles ready to publish?
The project says they generally need substantial editing.
Can it use a supplied document collection?
Yes. The examples include a vector-retrieval path for a custom corpus.
Can open-weight models be used?
Yes. The project provides an example, with setup and hardware requirements.
Does every citation prove its sentence?
No. Read the cited material and verify the connection.
Is the hosted preview a production service guarantee?
No. Treat research-preview access separately from a maintained deployment.
The details behind your decision
Explore platforms, inputs and outputs, licensing, and access requirements.
Four useful starting projects
Four useful starting projects
- An unfamiliar topic briefing: Start with a bounded question and use the generated outline to identify areas for further reading. Follow the sources for the central claims before presenting the briefing to colleagues.
- An editorial pre-writing packet: Collect background for an article, then turn the outline into an editorial plan. Separate verified facts, competing interpretations and unanswered questions before a writer creates the final narrative.
- A research-method demonstration: Compare a simple prompt with a structured research workflow on the same topic. Record sources and settings, and avoid treating one attractive draft as proof of generally superior performance.
- An internal knowledge prototype: Use a permitted document collection for a narrow organizational subject. Confirm that the retrieved passages match the question and that the model is allowed to receive those passages before sharing the resulting draft.
Platforms, inputs, outputs and setup
Platforms, inputs, outputs and setup
The Python package is named knowledge-storm. The repository documents separate research, outline, article and polishing stages, along with a minimal Streamlit interface. That interface displays intermediate progress and places an article alongside its references, which helps an editor revisit evidence.
For reproducible exploration, save the topic, model choices, retriever settings and source collection with the draft. Search results and model behavior can change over time. If an article appears comprehensive, still ask what the retrieval process failed to find and which sources were overrepresented.
Data, permissions and licensing
Data, permissions and licensing
A research topic can itself reveal a confidential plan. A self-hosted interface may still send queries and passages to external models or search providers, so map that route before using internal material.
The software license does not grant rights to every retrieved article or dataset. Retain source attribution and respect the terms of source material. When publishing, use your own analysis and concise supported summaries rather than reproducing long passages collected by the system.
Sources, documentation and help
Official sources and review date
Official sources and review date
Reviewed October 3, 2026. Product facts come from the official sources below. Suggested projects, prompts and review methods are FindGoodAI editorial guidance, not measured performance results.
Your experience can help someone choose
COMMUNITY
Sign in to rate and comment
Tried it? Share your experience
Loading experiences…
Manage my comments
Community and moderation
Saved tools are private. Approved comments are public; edits return to moderation. Comment counts include approved comments and replies. Each person has one rating. New ratings are automatically approved; withdrawn, unapproved or invalid ratings do not count. Share real experiences and avoid spam, private information or personal attacks.
For moderation appeals or data requests: support@findgoodai.com
NEXT STEP
Explore Stanford STORM ↗Build a stronger outline from traceable sources
Bring a real task and see how it fits the way you work.
Jump to a section
User experiences