# BondFoundry

> An MIT-licensed AI assistant for the buy-side fixed-income desk — and the working reference implementation of the FINOS AI Governance Framework v2.0. Self-hosted, Claude-backed, with hash-chained audit logs, HMAC-signed HITL approval envelopes, and a pure-function policy gate that produces verbatim rule citations on every blocked action.

BondFoundry sits alongside an existing OMS (it is *not* a replacement for BlackRock Aladdin, SimCorp Dimension, Charles River, or Bloomberg AIM). It is governed by a single pure-function policy gate, every action passes through that gate, every decision is written to a Postgres-enforced append-only audit chain with sha256 row chaining, and every audit row carries a `framework_ref` pointing to the FINOS AIGF v2.0 risk and mitigation it satisfies.

Coverage status: 23 of 23 AIGF v2.0 risks mapped. CI build fails any PR that drops coverage below 85%.

## Platform pillars

- [Policy gate](https://bondfoundry.esploro.group/platform/policy-gate): pure function `decide(action, context) -> Decision`. No I/O, deterministic, tier-routed by reversibility (T0 read-only, T1 reversible, T2 single-HITL, T3 dual-HITL + irreversible).
- [Audit chain](https://bondfoundry.esploro.group/platform/audit-chain): append-only at the database layer (Postgres triggers reject UPDATE/DELETE) with sha256 hash chain across rows for Tier-3 tamper-evidence.
- [HITL approval envelopes](https://bondfoundry.esploro.group/platform/hitl-envelopes): HMAC-signed, server-generated, scoped to action + identity. The agent cannot forge them.
- [AIGF coverage gate](https://bondfoundry.esploro.group/platform/aigf-coverage): CI-enforced 85% AIGF coverage minimum, generated by `bondfoundry-finos coverage`.

## Audiences

- [For the buy-side desk](https://bondfoundry.esploro.group/buy-side): operational velocity + materiality-tier governance for PMs, traders, treasury.
- [For engineers and CTOs](https://bondfoundry.esploro.group/engineers): architecture, MCP boundary, Postgres immutability, fastworker control plane.
- [For quants and AI architects](https://bondfoundry.esploro.group/quants): pure-function policy gate, four-dimension eval harness, model-agnostic design.
- [For FINOS and compliance reviewers](https://bondfoundry.esploro.group/finos): the reference implementation, cross-framework mapping, auditor evidence kit.

## Learn

- [How it works](https://bondfoundry.esploro.group/how-it-works): the governed-action lifecycle — agent proposes, the pure-function policy gate decides, HITL envelopes approve irreversible trades, every decision lands in a hash-chained audit log.
- [Use cases](https://bondfoundry.esploro.group/use-cases): the four governed-agent patterns — materiality-gated trade execution, tamper-evident audit for regulators, segregation of duties for AI agents, and a runnable FINOS AIGF v2.0 reference.
- [Compare](https://bondfoundry.esploro.group/compare): how BondFoundry relates to OMS platforms (runs alongside), agent frameworks (governance vs. orchestration), and vendor SaaS.
- [FAQ](https://bondfoundry.esploro.group/faq): governance, licensing, tier routing, the model boundary, and AIGF coverage.
- [Glossary](https://bondfoundry.esploro.group/glossary): definitions for policy gate, tier routing, HITL envelopes, hash chain, framework_ref, coverage gate, and more.

## Use cases

- [Materiality-gated trade execution](https://bondfoundry.esploro.group/use-cases/materiality-gated-trade-execution): let an AI agent draft fixed-income trades while every above-materiality order stops at an HMAC-signed HITL approval it cannot forge.
- [Tamper-evident audit for regulators](https://bondfoundry.esploro.group/use-cases/tamper-evident-audit-for-regulators): give auditors a hash-chained, Postgres-enforced audit log where every AI decision maps to a FINOS AIGF v2.0 control.
- [Segregation of duties for AI agents](https://bondfoundry.esploro.group/use-cases/segregation-of-duties-for-ai-agents): enforce four-eyes and dual-approval controls on an autonomous agent at the API layer — not the UI — so the agent can never self-approve.
- [FINOS AIGF reference implementation](https://bondfoundry.esploro.group/use-cases/finos-aigf-reference-implementation): adopt or study a working reference implementation of FINOS AIGF v2.0 with 23 of 23 risks mapped to code, evals, and audit rows.

## Articles

- [Blog index](https://bondfoundry.esploro.group/blog): all writing on auditable AI agents, materiality-tier routing, and the FINOS AIGF v2.0 control surface.
- [How to Build an AI Agent Your Auditor Will Actually Approve](https://bondfoundry.esploro.group/blog/auditor-approved-ai-agent)
- [The Four Pillars of Governed AI in Finance](https://bondfoundry.esploro.group/blog/four-pillars-governed-ai)
- [Materiality-Driven Access Control: Why Tier Routing Matters for Bond Desks](https://bondfoundry.esploro.group/blog/materiality-tier-routing)
- [Why Fixed-Income Desks Need FINOS AI Governance Now](https://bondfoundry.esploro.group/blog/finos-aigf-fixed-income)
- [Curve Provider Architecture for Multi-Source Bond Markets](https://bondfoundry.esploro.group/blog/curve-provider-architecture)
- [Evaluating Agentic AI: The Four-Dimension Battery](https://bondfoundry.esploro.group/blog/four-dimension-eval-battery)
- [FIX Gateway as AI: Connecting Your OMS to Claude-Backed Quant Analysis](https://bondfoundry.esploro.group/blog/fix-gateway-as-ai)
- [Open-Source vs Vendor SaaS: Building Your Own Governed AI Desk](https://bondfoundry.esploro.group/blog/open-source-vs-saas-ai-desk)

## Reference

- [About](https://bondfoundry.esploro.group/about) — what BondFoundry is and isn't, license, contributors
- [Book a walkthrough](https://bondfoundry.esploro.group/book) — 20-minute demo
- [Full-text dump for LLM ingestion](https://bondfoundry.esploro.group/llms-full.txt)
- [GitHub repository](https://github.com/neullabs/bondfoundry) — MIT-licensed source
- [RSS feed](https://bondfoundry.esploro.group/rss.xml)