Give IP Attorneys Patent Memory That Carries Every Prior Art Search
Each patent matter starts with a prior art search that overlaps 70% with the last one. AI tools forget every search the moment it ends. MemoryLake gives IP attorneys a patent memory layer — with built-in USPTO data and versioned prior art — so accumulated search work stops vanishing between matters.
Give IP Attorneys Patent Memory That Carries Every Prior Art Search
Get Started FreeFree forever · No credit card required
The problem: prior art research repeats itself every matter
The associate spent two days last month identifying relevant prior art in a specific subclass. This month, similar matter, same subclass — and the prior work isn't queryable. The AI tool starts from zero. Hours rebuild a search that already exists, costing the client and the firm.
How MemoryLake solves patent memory for IP attorneys
Built-in USPTO data
10M+ US patents queryable without separate setup. Add WIPO and EPO via REST connectors.
Per-matter and per-subclass memory
Prior art findings scoped to the matter and to the technical subclass.
Versioned claim analyses
Each claim chart commit becomes part of the matter's audit trail.
Conflict detection across precedent positions
Surfaces inconsistent claim interpretations before they reach a filing.
Free forever · No credit card required
How it works for IP attorneys
- Connect — Tap built-in USPTO data plus ingest internal matter docs, search histories, and claim charts.
- Structure — Each becomes typed memory: prior art (facts), strategy (reflection), filings (events).
- Reuse — Each new matter or claim analysis retrieves the relevant prior memory at the right scope.
Before vs. after: IP attorney patent memory
| Without MemoryLake | With MemoryLake | |
|---|---|---|
| New matter in a familiar subclass | Re-run prior art search | Retrieve prior search results |
| Claim chart consistency across filings | Manual cross-check | Conflict detection at write time |
| Associate turnover | Lose institutional searches | Memory stays at firm scope |
| Client audit "show your work" | Reconstruct manually | Versioned audit trail |
Who this is for
IP boutiques, in-house patent teams, and corporate IP groups using AI for prior art search, claim drafting, or matter management — where accumulated technical search memory across matters is a competitive advantage.
Try this use case on your own account
An IP firm keeps its prior art search memos in shared memory, one per US application, and its assistant answers from them through MemoryRouter: "what was the closest reference in our search for 17/482,907?" gets the right patent every time. Then an attorney asks about 17/428,907: two digits swapped, never searched. MemoryRouter attaches the memo for 17/482,907, and the model answers with that memo's reference under the number it was asked about. Nothing in the reply says the memo belongs to another application. Putting the list of searched numbers in the prompt stops most of these, but the model also "corrects" a number to the nearest one on the list and answers for it. An exact lookup in the app before any model call answers every near miss, and names the nearest docket number to check.
What you will see
- Six search memos imported into MemoryLake and the docket stored as six facts with
fact add;fact list --query 17/428,907returns 0 whilesearchranks the memo for 17/482,907 first, and--query 17/482907(no comma) also returns 0, so the app normalises numbers first - A checker (
router messages) shows what MemoryRouter attaches for a near-miss number: the neighbour's memo in 35 of 56 copies, nothing in 21 - Eight questions, 3 answers each through
router chat --boundary(gpt-4o-mini), over 7 runs: for the four near-miss numbers, memos only 0 of 84 "no search on file" (51 gave another application's answer), with the docket in the system prompt 73 of 84 (all 11 misses rewrote the number), with an exactfact list --querycheck before the model 84 of 84; the controls 250 of 252 - Plus
docketandmemossubcommands and a local web UI with the docket lookup next to what search returns
You need: a MemoryLake API key · the memorylake CLI · Python 3.9+
Run it
git clone https://github.com/memorylake-ai/memorylake-usecases.git cd memorylake-usecases/docket-number-memory export MEMORYLAKE_API_KEY=sk-… python3 demo.py # or: python3 web/server.py
Watch it run · real, unedited recordings
Related use cases
Frequently asked questions
Is client matter data isolated?
Is client matter data isolated?
Yes — per-matter namespaces with end-to-end AES-256 encryption. No cross-matter leakage possible.
Can we extend beyond USPTO?
Can we extend beyond USPTO?
Yes. WIPO, EPO, and any patent office with API or bulk export support imports through the SDK.
How does this handle privileged work product?
How does this handle privileged work product?
Privilege scopes apply at the memory namespace level. Audit trail records every access for malpractice review.