Architecture, benchmark methodology and results, model integration, the security model, and an honest read on compliance, for technical and security reviewers.
It sits beside your agents, never between an agent and its model: retrieval is a side call, not a new hop in the inference path. Add a source, every consumer benefits the same day.
Structured, cited, model-free. About 180 tokens. Handles most queries.
Accept, escalate, block, or abstain. Adds no facts.
Deterministic count / sum / list and per-entity comparison. Self-rejecting.
Reads only the already-selected document. Declines rather than fabricate.
True multi-hop questions across the selected document set.
The escalation tier is opt-in and off by default. Every escalation stays cited and is saved as draft knowledge for review, so no reasoning is wasted.
| Question type | Happy Knowledge | Strong RAG baseline | Context vs RAG |
|---|---|---|---|
| Factoid | 100% | ~70% | far less |
| Aggregation (count / sum) | 83% | 35% | up to ~133x less |
| Multi-entity comparison | 100% | ~85% | less |
| Multi-hop | 100% | 92% | less |
| Blended | 96% | 81% | ~22x less (median) |
Works with your Azure OpenAI deployment and key, so inference stays inside your existing Microsoft agreement and region. No vendor model in the path, no lock-in.
Consumed by people via plain-language chat, and by your AI agents via MCP and REST, both returning the same cited, governed facts.
Sources are re-compiled on change. A staleness checker flags artifacts due for review, so answers track your live policies.
Your own AWS account and region. Nothing is exfiltrated to train third-party models; no copy retained.
AES-256 encrypted keys, role-based access, sensitive-data detection, source lineage on every fact.
A full audit trail: who or what acted, under which authority, what was decided, and why.
Governance engine is Apache-2.0. Read the code and confirm the guarantees yourself.
Each agent runs inside an explicit authority boundary it cannot exceed. Trust expands by evidence.
Governance sits over the systems you already own, so adopting it does not widen your attack surface.
Candor is the differentiator. We tell you exactly where we stand so you make a real decision, not a reassured one.
Happy Technologies holds no SOC 2, FedRAMP, HITRUST, ISO 27001, or HIPAA Business Associate Agreement at this time. "Designed for regulated environments" is a statement of design intent, the way the architecture is built, not a certification we have earned. We say so plainly the day each one lands.
SOC 2, ISO 27001, and related attestations are sequenced work. Each marked the day it is real.
Architecture docs and security-posture detail are available today. Ask and we share.
The open-source governance engine is verifiable today. No attestation required. Verify, don't trust.
In its design-partner and pilot phase today; general availability in active development. Stated plainly, never implied as shipped.
A dedicated managed instance in your cloud. We run it; nothing to install. Live in days.
A fixed-scope pilot on a slice of your real corpus, with accuracy re-measured on your own documents before commitment.
Architecture and security detail on request. sales@happy-tech.biz