Stage 01
Ingest
Drop PDFs, scans, e-discovery productions, email batches. OCR and text extraction turn all of it into clean, cited text.
Inside Legalyzer
Four stages. One machine. Nothing in between.
The whole pipeline — extraction, indexing, analysis, recall — runs as local processes against local storage. Here is exactly what happens to a case file, and what it takes to run.
Stage 02
Index
Every page is chunked, vectorized, and stored in the local vector database. Ten thousand pages index overnight on minimum spec.
Stage 03
Analyze
Ask in plain language. Get page-cited summaries, chronologies, contradiction flags — inference runs on your GPU or Apple Silicon.
Stage 04
Recall
Notes are tracked per case and searchable across every engagement. New sources fetch, vectorize, and join the index on demand.
Architecture
Everything inside the line.
Your machine / your network — nothing crosses this line
Electron UI
Case workspace, reader, chat, notes
Local API
Every feature, scriptable on localhost
Inference engine
Local LLM on GPU / Apple Silicon; a lighter model handles extraction-only machines
DB
cases & notes
Vectors
document index
Cache
Redis-compatible
Sidecar Practice plan
Runs on the firm’s secure document server; your IT team manages it. Seats share one index over the LAN.
Connectors (opt-in, outbound only)
Clio · iManage · NetDocuments · Dropbox · Drive · OneDrive — fetched on request, indexed locally. Research links open in Westlaw / LexisNexis.
No telemetry. No account pings. License activation tolerates months offline.
System requirements
Sized for the hardware experts already own.
| Minimum — extraction, indexing & search | Recommended — full local LLM | |
|---|---|---|
| OS | macOS 13+ · Windows 11 x64 · Ubuntu 22.04+ | Same |
| Processor | Apple Silicon M1 · or 8-core x64 | M2 Pro or later · or NVIDIA RTX 4070+ (12 GB VRAM) |
| Memory | 16 GB | 32 GB |
| Storage | 20 GB free SSD | 60 GB free NVMe |
| What runs | OCR, indexing, search, summaries via the compact model | Full model — fastest summaries, deepest analysis |