Governed systems for industries that cannot guess
Law, banking, medical, accounting, and autonomous systems share one failure mode: an agent that acts without a gate, a receipt, or the right to refuse. We provide the software that belongs in that path — this month — and we say what is still research.
Why this is the safety and legal question
Post-hoc monitoring is not compliance. If the model has already allocated credit, altered a brief, or kept a patient identifier, the institution already holds the act. The gate has to refuse first. The ledger has to remember decisions, not chats.
Pre-effect refusal
A biased, injected, or low-coherence action is blocked before it executes. Logging it afterwards is not safety.
Receipts, not chat dumps
Every remember, refuse, and wipe can carry a Ghost Matrix or DGV hash. Counsel can replay a decision. They cannot replay a Slack thread.
Maps to regulation — does not impersonate a certificate
DGV cards align claims to EU AI Act, NIST AI RMF, ISO 42001, 21 CFR Part 11, and GDPR language. We do not sell SOC 2, HIPAA, or FDA clearance we have not obtained.
Human in the loop when the gate is thin
Abstention and HITL are first-class. An agent that cannot say no is not deployable in law, credit, or care.
What we can provide right away
Runnable APIs, crates, and verifier bundles. Limits are on the card. We will not sell a driving stack, an EHR, or a SOC 2 logo we do not have.
PrimeSwarm Cognitive API
Governed LLM proxy: sessions, tools, sandbox, JWT, PII redaction, fairness gate, cryptographic receipts. Helm chart included.
Unregulated agents create liability the moment they act. Governance has to sit in the execution path, not in a log after the fact.
- · HTTP API with streaming chat and tool calls
- · Pre-effect refusal on fairness / PII / sandbox violations
- · Receipted decisions for later audit
Limit: Requires an LLM key. Credit path is a gated demo, not a lending engine.
EMR Continuity Ledger
Durable agent memory: typed records, Ghost Matrix receipts, Ebbinghaus decay, abstention, fail-closed wipe. Chat dumps are never stored.
Regulators will ask what the agent remembered and why. A vector of transcripts is not an answer.
- · SQLite Continuity Ledger with WAL and audit JSONL
- · MCP tools: emr_remember, emr_recall, emr_upsert
- · Spatial XR wipe consolidates to DRAFT records only when requested
Limit: Phase 1 enforced crates. Not a clinical electronic medical record.
DGV Verifier
Deterministic & Governance Verified test cards and evidence receipts. Maps claims to EU AI Act, NIST AI RMF, ISO 42001, 21 CFR Part 11, and GDPR language.
Procurement and counsel need reproducible evidence, not a slide that says “we are compliant.”
- · Verifier bundle and scenario tests
- · SHA-256 evidence receipts
- · Claim registry for audit packets
Limit: Self-verification framework. Not a third-party SOC 2 or notified-body certificate.
Credit Advisor agent
Fairness-governed credit-decision assistant. Four-Fifths Rule and pre-effect refusal sit in the path, with receipts.
Fair-lending and employment decisions are where an ungoverned model becomes a lawsuit.
- · Adverse Impact Ratio check before execution
- · Human-in-the-loop when the gate refuses
- · Enterprise-tier marketplace listing
Limit: Decision-support gate, not an underwriting system of record.
Sandboxed code execution
Docker/WASM isolated execution for review and analysis agents. Used by Code Reviewer and Code Runner.
An agent that can run code without a box is an incident, not a feature.
- · Isolated tool execution behind the Cognitive API
- · SIEM-forwardable events
- · No host-network defaults
Limit: Isolation for lab and enterprise pilots. Not a full SOC-certified compute estate.
ONLY Lang + Trust Derivation Objects
Constraint language that heals, denies, or escalates. TDOs are replayable command traces for auditors.
Policy that cannot be replayed is not policy. A TDO is the receipt of the constraint.
- · Equilibrium scripts with fuel limits
- · TDO export for DGV / internal audit
- · Optional Groth16 path behind a feature flag
Limit: DSL for arithmetic and policy constraints. Not a general-purpose application language.
Only Forum research commons
Hosted research commons: abstract catalog, rooms, citation neighborhood, membership gate. Postgres is canonical speech.
Critical industries need a shared, sourced catalog — not a dump of PDFs into a chatbot.
- · Worldwide paper intake (abstracts only)
- · Identify-on-arrival and TPNN local map
- · Member writes after the free-view gate
Limit: Catalog and neighborhood, not a full-text RAG of every PDF.
Pilot or library
SignalO / PIR regime feed
Deterministic market-regime detection from return sequences. REST and live feed, alerts, historical replay. Not a trading bot.
Limit: Research / paper-trading product. Not licensed financial advice. Waitlist for hosted access.
PilotLearning Trails
Knowledge graph with PIR mastery scoring. Useful for internal certification programs once receipts are required.
Limit: Demo graph. Cryptographic mastery receipts are not wired yet.
PilotSpatial XR Dual-Front (Phase 1)
Deterministic routing of five agent cores, VISUAL_HANDOFF_V1, measured Think Mode, fail-closed wipe into the Continuity Ledger.
Limit: Core routing and wipe are enforced. The 3D render lane is declared, not shipped.
Critical & regulated
Banking & Markets
Regime detection, fair-lending gates, and audit-ready agent governance
Financial markets require deterministic signal generation, risk-adjusted position sizing, and immutable audit trails. Our PIR-based regime detection engine identifies market structural shifts in O(n) time, while PrimeSwarm ensures every agent decision is cryptographically receipted for compliance.
Healthcare
PII-safe agents, adversarial defense, and mastery-tracked medical education
Healthcare AI demands the highest standards of privacy, safety, and verifiable competence. PrimeSwarm's PII sanitization strips SSNs and NPI numbers before agent processing, TPNN prevents prompt injection from corrupting medical recommendations, and Learning Trails tracks structural mastery of medical knowledge — not just course completion.
Legal & Compliance
Immutable audit receipts, fairness gates, and zero-knowledge proof of compliance
Legal and regulatory workflows require provable fairness, immutable audit trails, and privacy-preserving verification. PrimeSwarm captures every agent decision with cryptographic receipts, fairness gates detect bias using the Four-Fifths Rule, and OnlyState enables zero-knowledge proofs of group membership without exposing individual identities.
Accounting & Audit
Replayable trade review, engagement memory, and receipted narratives
Close and attestation fail when the story cannot be tied to a hashed observation. PIR review pipelines and the Continuity Ledger give accountants a replayable narrative. We do not replace the general ledger or the statutory auditor.
Autonomous systems
A governance layer beside the vehicle or robot — not a driving stack
We do not ship perception, planning, or ISO 26262 hardware. We provide the AI decision gate: pre-effect refusal, coherence monitoring, injection defense, command routing, and a ledger that refuses to keep chat after a wipe. That is the part of the safety case counsel can actually read.
Commercial
Sales & CRM
Agentic inbox reading, company enrichment, and workflow automation
Sales teams spend hours on manual data entry, research, and follow-up scheduling. Minuta One CRM embeds durable AI agents directly into the sales workflow — reading team inboxes, enriching company data, and creating workflows autonomously. Agents maintain state across sessions, so research from yesterday continues today.
Creative & Media
Real-time coherence transport between music, visuals, and lighting tools
Live performances and installations require multiple creative tools to share structural state — not just MIDI clock sync, but verse/chorus, build/drop, tension/release. PCHI provides a conductor-based signal bus that routes coherence signals between Ableton Live, Resolume, and TouchDesigner in real-time.
Enterprise AI
Governed agent orchestration with 10-layer safety stack and streaming SSE
Enterprise AI deployments require governance, compliance, and reliability. PrimeSwarm provides a 10-layer governance stack — PII sanitization, adversarial defense, fairness gates, human-in-the-loop approval, cryptographic receipts, SIEM integration, JWT auth, GDPR retention, hardened sandboxing, and MCP tool protocol. No competing framework offers more than 3 of these.
Infrastructure & security
Web3 & Blockchain
O(1) social state, zero-knowledge proofs, and governed on-chain agents
Blockchain applications face state bloat, privacy tradeoffs, and ungoverned agent interactions. OnlyState compresses social graphs into constant-size on-chain state with ZK-PIR, PrimeSwarm governs on-chain agent execution with cryptographic receipts, and Quilt-AGI Bridge provides ZK proofs for cellular automata evolution.
Cybersecurity
Mathematical adversarial defense, PII sanitization, and SIEM-ready audit trails
Security teams need AI that cannot be hijacked. TPNN provides mathematical guarantees against prompt injection — attacks outside the topological manifold are physically incapable of forming. PrimeSwarm strips PII before agent processing and forwards all events to SIEM systems in real-time.