Repository navigation
Heimdall v1.0.0 — IC-Grade Intelligence Platform
First major release. Heimdall now produces analytic products that conform to authentic Intelligence Community formatting, sourcing, and tradecraft standards (ICD 203), with cross-referenced threat indicators and analyst-grade structured analytic techniques.
What's new
📋 Multi-format reports (NIE / PDB / IIR / Full Assessment)
Pick the output format from a dropdown in the chat toolbar — or let it auto-select based on your query. Each format mirrors the structure of authentic declassified products (ODNI ICAs, NIEs, DIA IIRs, Nixon/Ford-era PDBs).
🎯 ICD 203 compliance scoring + auto-regeneration
Every report is scored against 8 of the 9 IC analytic tradecraft standards. Reports below 70/100 trigger a one-shot LLM regeneration with deficiency notes. The full Words of Estimative Probability scale (almost no chance → almost certainly) is now mandatory in analyst output.
🛡️ Threat-feed cross-reference
- MITRE ATT&CK: 1,647 indicators (172 actors / 693 malware families / 91 tools / 691 TTPs) auto-ingested on first boot, weekly refresh
- MISP public feeds: CIRCL OSINT, botvrij, ThreatFox available with daily refresh
- Every IOC mentioned in your reports gets cross-referenced — known-bad indicators are annotated with severity icons (🔴/🟠/🟡)
- Dark-web report severity now auto-bumps when content matches threat-feed indicators
📚 CIA CREST training corpus + few-shot exemplars
- Pull declassified IC documents from Internet Archive's 851,358-doc CIA mirror by topic
- Pre-OCR'd full text — no PDF parsing required locally
- ExemplarSelector picks topic-matched real declassified docs and injects them into the analyst prompt as format references
- Era-bucketed: cold_war / post_cold_war / gwot / modern
🧩 Structured Analytic Techniques (4 SATs)
Toggleable from the chat toolbar:
- ACH — Analysis of Competing Hypotheses (Heuer matrix with C/I/NA scoring, leading hypothesis identification)
- KAC — Key Assumptions Check (fragility rating + verifiability)
- RedT — Red Team (Devil's Advocacy + Red Hat adversary perspective)
- I&W — Indicators & Warnings framework (confirming/refuting indicators per key judgment)
Each appends a typed annex to the report; results visible directly in the chat output.
⚙️ Settings → Training Data tab
- Total indicators / per-source breakdowns / last-sync times
- Manual MITRE/MISP/CREST sync controls
- Live IOC scanner (paste text → see threat-feed matches)
- Tradecraft Compliance History (avg score, pass rate, regen count, scrollable list of recent reports)
🧱 8 new workflow nodes
Drop into custom workflows in the visual editor:
FormatSelector · ACHMatrix · KeyAssumptions · RedTeam · Indicators · ThreatFeedCheck · TradecraftScore · ExemplarInject
🗄️ Database
Migration 041 adds three new tables:
training_corpus— declassified docs for few-shot exemplarsthreat_feeds— MITRE/MISP indicators with stix_id and misp_event_id linkingreport_quality_scores— ICD 203 history per report
Verification
16 unit tests (vitest) covering ThreatFeedMatcher and MispFeedIngester. Run:
npm testDownloads
- macOS Apple Silicon (arm64):
Heimdall-1.0.0-arm64.dmg - macOS Intel (x64):
Heimdall-1.0.0.dmg
Both unsigned — when first opening, right-click → Open to bypass Gatekeeper, or xattr -dr com.apple.quarantine /Applications/Heimdall.app.