Defensive Publications · Dated Prior Art

IP Publications

Open technical disclosures (defensive publications) from Gus IT LLC — published as dated prior art and reference open-source implementations.

⚖️ License & prior-art notice

These disclosures are published as public prior art: a dated, citable description of each technique, intended to establish the state of the art. The technique itself is freely practiceable by anyone.

The accompanying reference code is licensed AGPL-3.0-or-later (AGPL-3.0-or-later) — a copyleft licence: anyone who modifies it and offers it as a network service must release their corresponding source. A commercial licence is available from Gus IT LLC for proprietary or closed-source use.

For a commercial licence or any licensing question, contact [email protected].

Tier 3

Proactive-Action Engine with Layered Cooldowns

Idle-time multi-source agent scanner with pre-LLM role gates and per-action-type layered cooldowns.

Tier 3

Multi-Provider Voice Pipeline with Lazy Key Routing

Real-time STT→LLM→TTS pipeline with call-time per-persona provider selection, lazy bridge-resolved keys, and three-stage latency telemetry.

Tier 2

Persona Mailbox Auto-Provisioning with Credential Rehydration

Per-agent JMAP mailbox auto-provisioning with two-id resolution and boot-time credential rehydration.

Tier 2

Dynamic Confidence Bands per Action Type

Per-action-type confidence bands with rolling, human-free recalibration and an auto/escalate/review routing trichotomy.

Tier 3

Unified Multi-Channel Messaging History

Single-table JSONB-metadata design with a canonical-contact timeline for cross-channel agent message history.

Tier 3

Hybrid RAG Ranking with Graceful LLM Fallback

Two-stage RAG ranker whose optional LLM rerank stage silently fails open to a deterministic base.

Tier 2

Two-Tier Attention Queue with Amortized Triage

Bounded in-memory hot tier overflowing to an unbounded persistent cold tier, with an exactly-once triage invariant.

Tier 2

Resource-Provider RBAC for Multi-Domain AI Platforms

Boot-time, resolver-invariant resource-provider RBAC with path-scoped wildcard tuples and AI-persona principals.

Tier 2

Per-Persona Token Budget Tracker with Strategy Bias

A cost-as-control-input budget signal that biases an LLM agent's strategy selection toward frugal alternatives.

Tier 3

Idle-Triggered Curiosity Engine for AI Agents

Idle-triggered, weight/cooldown/availability-gated curiosity actions selected by weighted-random draw.

Tier 3

Capability-Scoped Task Envelopes with Reserved-File Token Broker

Capability-scoped task envelopes (dispatch≡merge identity) plus a reserved-file token broker for concurrent AI coding agents.

Tier 3

Task-as-Hub Coordination for a Mixed Human/AI-Agent Workforce

A DB-backed Task record owns a unit of work and arbitrates concurrent human/AI-agent writes in lieu of git branches.

Tier 3

Per-Route Fail-Closed Multi-Provider LLM Routing

Fail-closed-vs-open routing bound to the semantic route class, with audit and a self-degrading fallback chain.

Tier 3

Tiered-Boot Capability Kernel for an AI-Agent Platform

A microkernel tiered-boot assembler whose Tier-0 is operable and health-serving before any AI tier loads.

Public GitHub repositories for each disclosure are published under the gusitllc organization.