Upload files and scans, then ask questions in plain language. NeedleSearch searches your whole archive, checks every answer against its sources, and can run securely on your own servers or through an API.
A data room should not become a scavenger hunt. Upload it once and ask a plain-language question. NeedleSearch searches the full collection, connects the relevant facts and returns an answer you can check against the original page.
Contracts, scans, PDFs and tables, including image-based files — plus hearings, calls and dictation, transcribed. Encrypted from the moment they arrive.
No query language. Write the question as you'd ask a colleague, and choose how deep the check should go.
Each finding names its document and page. Preview the passage, then open the source in one click.
No special query language. Just your documents, your question and a clear trail back to the evidence.
Upload contracts, scans, PDFs, tables and more. NeedleSearch reads every page, then organises the collection so it can be searched by meaning — not just by an exact word.
Write it as you would say it to a colleague. An agent breaks the question down and hires sub-agents to search the collection and re-check the facts, then writes a direct answer with the exact page behind it. Answers come only from your documents.
Each finding names the document and page it came from. Preview the original passage, then open the full document in one click.
Six jobs that used to take days of reading. Each one starts the same way: upload the files and ask the question below in your own words.
Thousands of contracts, corporate records and financials. NeedleSearch finds change-of-control clauses, missing consents and documents that contradict each other, with the page for each.
Letters, pleadings, minutes and witness statements become one dated timeline, every event with its source. Where two documents tell the story differently, you see both.
A word-by-word comparison of two versions, then what each change means for your side. Every obligation and deadline goes into one list: who, what, by when.
Match line items between estimates, invoices and payment records, even when the columns are named differently. Every mismatch is listed with its row and size, largest first.
Before files go to a court, regulator or counterparty, find every name, passport number, bank detail and confidential passage, including on scanned pages.
Search European Court of Human Rights case law, or your firm's own memos and opinions, by meaning rather than exact keywords.
Finding the page is the first step. NeedleSearch also does the work that usually follows it.
Ask the same questions of every file (party, date, amount, governing law) and get one spreadsheet back, each cell linked to its page.
Upload Excel or CSV files with hundreds of thousands of rows. Totals, counts and filters are calculated exactly, not estimated from a sample.
A precise redline of two versions, or several documents checked against each other for names, dates, amounts and defined terms.
Deadlines, interest and totals are computed, never guessed, including working days under the holiday calendars of 251 jurisdictions.
Long reports written section by section, with diagrams and timelines where they help. Export to Word or PDF.
A company register, a statute in force, an exchange rate: when a document states an outside fact, NeedleSearch can check it on the original website.
A skill is a written procedure the agent follows step by step, the way an experienced associate runs a due-diligence review or an audit. NeedleSearch picks the right one for your question.
Your team can add its own: write your house method once, in plain English, and every answer in your workspace follows it.
Goes through a deal's documents in order: first checks that nothing is missing, then lists the risks it found.
Same folders. Same questions. Same verified answers, checked against the original passage — because a benchmark is only useful when the evidence holds up.
In plain terms: where your files are kept, who can see them, and what we never do with them.
Files are stored on our own servers in Sweden and encrypted as soon as they arrive. The keys are held in a separate key service, not next to the data.
The database itself keeps each customer's data apart, so one account cannot reach another's. The wall is not just a check in the app.
Personal files, files private to a team and files shared with the whole team. Colleagues see only what you share with them.
We do not use your documents to train, fine-tune or evaluate our models.
Delete a file and it is removed from the database, the search index and storage. Backups roll over within 30 days.
If your documents cannot leave your network, NeedleSearch can run inside your own infrastructure.
Everything in the app is available through a REST API and an MCP server. Connect NeedleSearch to Claude, ChatGPT or Cursor, or call it from your own code with the Python SDK.
# pip install needlesearchai from needlesearchai import NeedleSearch ns = NeedleSearch(api_key="nsk_...") results = ns.search("termination clauses") job = ns.jobs.create("Summarise the risks", item_ids=[doc_id]) print(ns.jobs.wait(job.id).answer)
Every plan reads your whole archive and cites every answer. Moving up raises the limits — usage, concurrency, storage — rather than unlocking the basics.
No. We do not use your documents to train, fine-tune or evaluate our models.
On our own servers in Sweden, encrypted at rest. If your documents cannot leave your network, NeedleSearch can run on your own servers instead.
Not yet. We are preparing for both certifications and are putting the required security policies and controls in place now.
Word, PDF, PowerPoint, Excel and CSV, emails, scanned pages and photos, audio recordings, web pages and archives of these. Scans are read with text recognition, and recordings are transcribed.
Tens of thousands of files; we test on over 30,000. In very large collections a quick search may miss something, so use the deeper search mode.
Every claim names the document and page it came from. Hover to preview the passage, or open the original in one click. If the documents do not contain the answer, NeedleSearch says so instead of guessing.
Yes. A team shares one workspace, with personal folders, folders private to a group and folders shared with everyone.
Yes. Every plan includes the REST API and the MCP server, so NeedleSearch works from Claude, ChatGPT, Cursor or your own code.
We taught the search agent to tell a lexical query from a semantic one. Case numbers and article references are now found in 99.8% of cases instead of 48.6%, and empty results dropped from 32% to 11%.
Why we built a calculation tool instead of letting the model do the sums, how it handles working days across 251 jurisdictions, and a measured comparison: 64% of answers correct without it, 100% with it.
One question. The right evidence. A clear way to check it.