⌁
Extract structure
Parse programs, jobs, copybooks, screens, tables, triggers, and files locally—with no model and no tokens.
LAYER 1 · DETERMINISTIC Open source · Model agnostic · Apache 2.0
Legacy knowledge.
Made durable.
Turn COBOL, JCL, CICS, DB2, and VSAM estates into a portable knowledge graph your team owns—then spend AI tokens only on logic that actually changed.
Zerotokens to extract
5artifact families
100%portable data
JCL JOBPAYRUN
PROGRAMINTCALC01
DATASETACCOUNTS
PARAGRAPHCALC-INTEREST
TABLETXNLOG
COPYBOOKACCTFLDS
graph ready
impact of ACCOUNTS△ 0.92High-risk logic foundQueued for review · not reprocessed
COBOL ✦ JCL ✦ CICS BMS ✦ DB2 ✦ VSAM ✦ SQL PL ✦ KNOWLEDGE GRAPHS ✦ ZERO LOCK-IN
01 / THE SYSTEMUnderstanding that
Understanding that
compounds.
Deterministic machinery handles facts. Models handle ambiguity. Every verified insight persists, so your knowledge base gets richer while recurring cost approaches zero.
◇
Build the graph
Connect calls, reads, writes, executions, and lineage in a versioned SQLite graph you can inspect and export.
LAYER 2–3 · PORTABLE✦
Enrich selectively
Redact sensitive data, rank risk, and narrate only new or changed business logic using the model you choose.
LAYER 3.5–4 · CONTROLLED↗
Query impact
Answer “what runs this?” and “what breaks if this changes?” through instant, token-free graph traversal.
LAYER 5–6 · ACTIONABLE02 / BUILT DIFFERENTOwn the intelligence.
Own the intelligence.
Not the invoice.
Your source stays local by default. Your graph is a standard, versionable artifact. Your model provider is replaceable.
CONTENT HASHING
Pay once for understanding.
Post-expansion, post-redaction hashes mean unchanged paragraphs hit the cache forever. One changed line re-queues one unit—not an entire application.
2000-CALC-INTEREST
987b31a8CACHED9000-ERROR
7b1c624eCACHED3100-POST-TRANSACTION
new contentQUEUESensitive data stops here.
Credentials and financial identifiers are scrubbed before any optional model call.
ROUTING-NO = 021000021
ROUTING-NO = [REDACTED]
ROUTING-NO = [REDACTED]
Trust, with provenance.
Verify, flag, edit, and stale-check business rules. Every insight remains traceable to source.
✓ Verified⚑ SME review
03 / GET STARTEDFrom source to insight
From source to insight
in three commands.
Runs offline with the built-in mock adapter. Connect Anthropic or OpenAI only when you are ready.
Read the full guide ↗QUICK START
# Install from source
pip install -e ".[dev]"
# Extract structural facts
mainframe-brain extract examples/cobol \
--out brain.db
# Ask impact questions—zero tokens
mainframe-brain query --store-path brain.db \
"impact of ACCOUNTS" Local-first · No API key required
04 / SUPPORT THE PROJECTKeep legacy
Keep legacy
knowledge free.
Mainframe Brain is independent and open source under Apache 2.0. Sponsorship funds parser coverage, documentation, and the long-term maintenance that keeps this knowledge in your hands—not behind a paywall.
Fund the roadmap.
Recurring or one-time sponsorship goes directly toward development. Every tier helps sustain the project and unlocks a place in our supporters list.
Become a sponsor ♥★
Star the repo
A star boosts visibility and helps other mainframe teams discover the project.
Star on GitHub ↗⌁
Contribute
Add parsers, fix bugs, or improve docs. Good first issues are labeled and welcome.
Contributing guide ↗✦
Spread the word
Share with your team or write about your experience. Word of mouth keeps open source alive.
Join discussions ↗THE KNOWLEDGE SHOULDN'T RETIREGive your mainframe
Give your mainframe
a memory.
Open source. No black box. No forced migration.
Explore on GitHub ↗