You have a case
The discovery requests are out. The documents are coming. Where do you put everything?
SharePoint, a review platform, Google Drive, or Geistra. They'll all hold the files. If you want AI to work on your case, Geistra is the answer.
SharePoint
Your firm's files, in folders.
A review platform
Productions, coded by reviewers.
Google Drive
Files and docs, in folders.
Geistra Legal
Your case, ready for your AI.
Built for AIGeistra is a data harness built for AI, in the age of AI.
It reads every document, hands your AI exactly what it needs, and keeps what it works out for your whole team. Here's the same case, three ways.
Step 1 · Upload the discovery requests
Turn the requests into something you can track.
You ask“Track these requests as documents come into the case.”
SharePoint
There's nothing to ask. The requests stay a file.
Google Drive
There's nothing to ask. The requests stay a file.
Geistra Legal
| Request | Status |
|---|---|
| RFP 1: Communications with the HOA | Waiting |
| RFP 4: Board minutes, 2019 to 2023 | Waiting |
| RFP 7: Gate installation invoices | Waiting |
| RFP 12: Insurance policies | Waiting |
| RFP 15: Photographs of the gate | Waiting |
Every request becomes a line your team can track. Ready before the first document arrives.
Step 2 · The documents pour in
Lots of them. Can your AI actually read them?
They'll all store them. The question is how much of each one your AI can ever search. Here's a 2,000-page production, in one PDF:
Scans are read
Geistra reads every scanned page. SharePoint's search doesn't read scans; Google Drive's stops at page 80.
Big files are read
Every file up to 2 GB, in full. SharePoint's search skips anything past 150 MB (512 MB for PDF and Office); Google Drive's, anything past 10 to 50 MB.
Zips are opened
Upload a whole production as one zip and every document inside is read. SharePoint and Google Drive store it as one file.
Step 3 · Match documents to the requests
Your AI does the matching. Where does the work go?
You ask“Match the production to the requests. What's missing?”
SharePoint
Here's what I matched… (a long chat reply)
The answer lives in your chat. Close it and it's gone. Your colleague can't see it, and their AI starts over.
Google Drive
Here's what I matched… (a long chat reply)
Same: it stays in the chat. Nothing on the list of requests changes.
Geistra Legal
| Request | Status |
|---|---|
| RFP 1: Communications with the HOA | Received |
| RFP 4: Board minutes, 2019 to 2023 | Received |
| RFP 7: Gate installation invoices | 1 missing |
| RFP 12: Insurance policies | Received |
| RFP 15: Photographs of the gate | Missing |
Each request now points to its documents. The list is up to date, for the whole team.
Step 4 · Work the case
Now ask the questions that matter.
You ask“Which witnesses said the gate went in without an HOA vote?”
You ask“Compare the depositions and highlight where the witnesses contradict each other.”
Your AI can only take in so much at once. Think of a mail slot. What comes back decides whether your AI sees all the testimony, or only some of it.
The mail slot
Your AI has a mail slot. Only so much fits through at once.
What you push through it decides what your AI knows.
SharePoint or Google Drive
Your AI answers from the part that got through.
Geistra Legal
Every passage, from every transcript, gets through.
The HOA vote question
The testimony is on page 188. Does your AI ever get there?
“Which witnesses said the gate went in without an HOA vote?”
SharePoint or Google Drive
The transcript is handed over whole. Your AI reads the first part, runs out of room, and answers from that.
Geistra Legal
Every page was read when the transcript arrived, so the search goes straight to the page that answers.
Your AI gets page 188, not pages 1 to 80. From every transcript.
Comparing the depositions
Same question. Count the trips.
“Compare the depositions and highlight where the witnesses contradict each other.”
SharePoint or Google Drive
Ends with: part of the testimony
- Search
- List of files
- Fetch transcript 1 ✂
- Fetch transcript 2 ✂
- Fetch a scan: blank
- Fetch transcript 4 ✂
- Out of room
✂ = cut off: too big to read in full. Every trip fills the chat, until there's no room left to compare.
Geistra Legal
Ends with: all the testimony, side by side
- Search
- The testimony, with pages
- next
- next
- ✓ Compared
No documents fetched. Scanned exhibits included. Room left over for the actual work.
What your AI actually sees
Look over your AI's shoulder.
The same question, the same AI. Here's what actually comes back to it.
Your AI, connected to SharePoint or Google Drive
Which witnesses said the gate went in without an HOA vote?
search → a list of files
read Whitlock deposition.pdf → too large to show, saved to a file instead
read HOA minutes (scan).pdf → no text
read Quinn deposition.pdf → …
“Based on the documents I could review, it appears…”
Your AI, connected to Geistra Legal
Which witnesses said the gate went in without an HOA vote?
search → the testimony on the gate, with pages · 1 of 3 · next
next → 2 of 3
next → 3 of 3 · done
“Three witnesses: Quinn, installed without an HOA vote (Quinn Dep. 112:4–19); Whitlock, the board approved it by email (Whitlock Dep. 188:4–19); Harlan, never told of a vote (Harlan Dep. 41:2–9).”
What one answer an AI app reads, per Anthropic's documentation: about 150,000 characters on claude.ai, 50,000 in Claude Code (longer answers are saved to a file instead). A 300-page document is about 350,000. Geistra Legal sends at most 30,000 at a time. OpenAI doesn't publish ChatGPT's limit.
Step 5 · Keep going
Every answer is kept. Every next question is faster.
In Geistra, what your AI finds is kept as data your team can filter, update and build on. Tomorrow your colleague asks the same question and their AI already knows. Ask for a dashboard and it's there, and it stays current.
Less time
The next question starts from what's already found.
Fewer tokens
Your AI reads the findings, not the transcripts again.
More accurate
Every answer points to its page, so you can check it before you rely on it.
Case dashboard · discovery
Live · updated when the supplemental production arrivedRequests tracked
22
Gaps
5
Waiting on the other side
4
RFP 7 invoices
1 missing → Received
closed by the new production
With SharePoint or Google Drive, every new chat starts from zero and reads the transcripts again, and a dashboard is a document that's out of date by Friday.
Pay your AI to re‑read, or to answer.
60 days of discovery. 300 questions from the case team. Here's what Claude reads to answer them.
An estimate, not a live run: 4,000 pages, 5 questions a day. Claude's input tokens at Claude Opus 5.5's API price ($4 per million). On a Claude plan, the same tokens count against your plan's usage limits.
SharePoint and Google Drive keep your files. Geistra Legal lets you work your case with AI.
30 minutes. Your own documents, in Claude or ChatGPT.
Book a demoLimits from Microsoft's, Google's and Anthropic's own documentation: Search limits for SharePoint (Microsoft); SharePoint limits (Microsoft); Google Drive indexing limits (Google); Tool result size limits in Claude (Anthropic); MCP output limits in Claude Code (Anthropic). Products change; check there for the latest. The case, witnesses and documents shown are fictional, and the answers are illustrations. SharePoint and Microsoft 365 are trademarks of Microsoft; Google Drive is a trademark of Google; Claude is a trademark of Anthropic. Geistra is not affiliated with or endorsed by any of them.