Technical Reference & SDK Specifications

Ironclad Grants & @tekromancy/grant_utils Docs

Complete architectural documentation for PostgreSQL pgvector HNSW RAG grounding, 2 CFR 200 statutory math engines, decentralized 24GB+ Billama GPU routing, RFC 5545 calendar feeds, and Kubernetes Helm deployment.

Open Interactive Docs in Portal

v2.4.0 • 185 Tests

Documentation Modules (8)

Core Architecture

Grant Operating System Architecture

End-to-end architecture of Ironclad Grants: 3-phase proposal lifecycle, PostgreSQL pgvector HNSW RAG, SeaweedFS object storage, FreeIPA LDAP, and 24GB+ decentralized Billama GPU inference.

Try Live in Portal

Ironclad Grants is an enterprise, multi-tenant **Grant Operating System (Grant OS)** engineered for 501(c)(3) nonprofits, worker cooperatives, universities (Offices of Sponsored Programs), and municipal agencies.

Three-Phase Lifecycle + Post-Award OS - Phase 1: Funder Intelligence & 990-PF Hydration: Synchronizes federal opportunities (Grants.gov, DOE, USDA, NIH, NSF, EPA) and private foundation tax filings via ProPublica Form 990-PF auto-hydration. Evaluates 5 dimensions (Mission Alignment, Legal Eligibility, Funding Size Fit, Cost-Share Feasibility, Competitiveness) to compute statistical ask "sweet spots" and letter grades (`A+` to `F`). - Phase 2: Institutional Knowledge Base & HNSW RAG: Stores organizational 501(c)(3) determination letters, audited Form 990s, biosketches, and board rosters inside isolated Supabase PostgreSQL (`pgvector` HNSW) and encrypted SeaweedFS vaults. - Phase 3: AI Copilot Studio & Red Team Review: Routes 8-section proposal synthesis through 24GB+ decentralized Billama GPU nodes via `@billama/sdk`, enforces source-level RAG citations, runs a 3-persona simulated peer review (1–9 scale), and packages a pre-flight validated `.ZIP` Submission Binder. - Phase 4: Post-Award 2 CFR 200 Financial OS: Compiles OMB SF-425 Federal Financial Reports, solves 2 CFR 200.414 Modified Total Direct Costs (MTDC), and tracks 2 CFR 200.430 personnel effort variances.

Multi-Tenant Isolation & Zero-Training Guarantee | Layer | Technology | Security & Isolation Mechanism | | :--- | :--- | :--- | | Database & Vector Index | PostgreSQL 16 + `pgvector` HNSW | Row-Level Security (RLS) enforced at kernel level by `org_id` | | Directory & Auth | FreeIPA LDAP + Supabase Auth | RBAC across `owner`, `admin`, `writer`, `reviewer`, and `guest` | | Artifact Vault | SeaweedFS Distributed Object Store | Encrypted buckets for IRS letters, SF-424A CSVs & ZIP binders | | AI Inference | Billama Decentralized 24GB+ GPUs | Ephemeral inference with strict Zero Foundation Model Training | | Audit Trail | SHA-256 Hash Chain | Tamper-evident cryptographic chain for Single Audit defense |

Executable Code & CLI Reference

Install Core Engine & Billama SDK

pnpm add @tekromancy/grant_utils @billama/sdk

Initialize Multi-Tenant Store & Evaluate Readiness

import {
  createSupabaseStore,
  evaluateGrantReadiness,
  getKPISummary,
  getAllGrants,
} from '@tekromancy/grant_utils';

const store = createSupabaseStore();
const grants = getAllGrants();
const kpi = getKPISummary(grants, 4500000);

console.log(`Active Pipeline: $${kpi.totalPipelineValue.toLocaleString()} (${kpi.activeGrantsCount} grants)`);