parsecache
Get a demo
[ Intelligent DMS layer ]

Turn your DMS documents agent-ready.

A read-only layer over IBM FileNet, Documentum, and OpenText. Every document becomes agent-ready markdown and artifacts, parsed once and cached so you never pay to re-OCR.

Get a demoSee how it parses
parsecache console
IBM FileNet P8Connected
Sourcefilenet-p8://cpe-prod:9080/FileNet/prod-1arov1
Requests / min
0
▲ 4.2% vs avg
Cache hit rate
0.0%
served from cache
Avg completion
0.0s
p50
Time in queue
0.0s
p50
Throughput, last 20 minlive
req/mincache hit %
Saved this month
~$0.0k
[ Live parse queue ]0 queued
/v1/documents
DocumentEngineQueueParseStatus
KYC{A17E40C2…9D4}paddle-ocr-vl-1.66.1sparsing
Wire{6B31D0A4…F07}cache0.0s6msCACHE HIT
Deed{C4F91E77…3AC}gemini-3.5-flash4.4sparsing
Mortgage{2F5B9C10…B21}cache0.0s8msCACHE HIT
AML{90AD7B52…1E8}paddle-ocr-vl-1.61.8sparsing
[ Architecture ]

A read-only layer over your DMS.

A request comes in, parsecache reads the document straight from your DMS, and returns clean markdown.

request
GET /documents→
markdown
✓
# Master Service Agreement
**Parties.** Northwind Capital LLC
## 3. Fees
Total $1,240,000.00
Master Service Agreement
Master Service Agreement
parsecache
Extraction engine

Route each element to the right engine.

Tables go to OCR for exact cell boundaries. Signatures go to a classifier. Clauses go to a VLM. parsecache runs them per element and caches the combined result.

Master Service Agreement
Authorized signatory
Extraction output
99.2% accuracy

The recommended pipeline. Best extractor per element.

FieldData
table.line_items
scanning…
total_amount
scanning…
signature
scanning…
effective_date
scanning…
governing_law
scanning…
0/5 elements, mixedparsing…
[ Intelligent layer ]

A layer on top of your DMS.

It powers agentic automation over documents your models otherwise can't read. Trigger it on upload or backfill in bulk, and your agents and workflows get fast, structured access to every document in a format they can act on.

Uploads
IBM FileNet
parsecache
Agents
New uploads are parsed the moment they land in the DMS, then served to your agents.
01
Parse on upload
a webhook fires the moment a document lands in the DMS
02
Backfill the store
sweep every existing document once, then keep it warm
03
Feed your agents
hand agents and workflows the parsed document over one API
[ Output ]

One document in. Markdown + artifacts out.

Every document is exported as clean markdown text. Non-text elements like figures, diagrams, signatures, and stamps are lifted out and saved as separate artifacts, each referenced inline.

Source document
Master Service Agreement
loan_agreement.pdf, 6 pp
parse▶
Parse resultmarkdown
{2F5B9C10…B21}.md
# Master Service Agreement


**Parties.** Northwind Capital LLC and Vertex Holdings Inc.,
effective 2026-06-01.


## 3. Fees


![fig_1: fee schedule](artifacts/fig_1_table.png)


Total consideration is $1,240,000.00, payable per the schedule above.


## 9. Governing law


Governed by the laws of the State of New York.


![fig_2: entity structure](artifacts/fig_2_diagram.svg)


![sig_1: authorized signatory](artifacts/sig_1.png)
![stamp_1: corporate seal](artifacts/stamp_1.png)
Extracted artifacts→ cache
fig_1_table.pngfigure0.99
fig_2_diagram.svgdiagram0.97
sig_1.pngsignature0.986
stamp_1.pngstamp0.94
Any provider

Bring your own OCR and VLM.

Map each element type to an engine. Managed APIs or self-hosted models. parsecache normalizes the output to one schema and caches it.

Paddle-OCR-VL 1.6Gemini 3.5 FlashAWS TextractAzure Document IntelligenceGoogle Document AIGPT-4o VisionClaude VisionSurya OCRdocTRPaddle-OCR-VL 1.6Gemini 3.5 FlashAWS TextractAzure Document IntelligenceGoogle Document AIGPT-4o VisionClaude VisionSurya OCRdocTR
Paddle-OCR-VL 1.6Gemini 3.5 FlashAWS TextractAzure Document IntelligenceGoogle Document AIGPT-4o VisionClaude VisionSurya OCRdocTRPaddle-OCR-VL 1.6Gemini 3.5 FlashAWS TextractAzure Document IntelligenceGoogle Document AIGPT-4o VisionClaude VisionSurya OCRdocTR
One normalized schema across every engine.
Pipeline routing
per element
Tablespixel-perfect grids
Signaturesink-color + intent
ClausesVLM reasoning
Everything elsefallback
Cachestoreparsed-docsretention
[ SDK suite ]

A typed SDK suite for every language.

Fetch a parsed document by its DMS id. The first call parses, every call after is a cache read. Same request shape across all of them.

Python
TypeScript
Java
.NET
Go
REST
Business impact

Parse once. Every read after is a cache hit.

The first request parses and writes to the cache. Every later read returns the stored result instead of re-running OCR, for any team or service. No repeat cost, no extra load on the DMS.

0%
lower OCR spend
Re-reads hit the cache at $0. You pay the provider once per document, ever.
0 container
Simple deployment
One read-only sidecar next to FileNet. Point it at a bucket. No migration, no writes.
0×
faster iteration
Tune a downstream workflow against cached parses. No re-OCR, no per-run cost.
Savings scale with your cache hit rate.
Plug in your page volume and OCR pricing to model it on your own numbers.
Estimate your savings

Parse once.
Serve every read from cache.

One read-only sidecar in your VPC. No data leaves your network, no writes to the DMS, no per-read OCR bill.

Get a demoRead the docs