Software and research code.

The full reference implementation, including Subjective Logic, FHIR interoperability, security hardening, temporal operations, and AI/ML metadata.

A feature-complete JavaScript/TypeScript implementation with built-in type declarations and an MCP server.

The formal gap analysis and extension proposal behind the framework.

Normalized token accounting, effective-budget gauges, turns-to-exhaustion prediction, and compaction/handoff policies.

Terminal gauges and renderers that present context-budget telemetry.

Language-parity implementations and a unified handoff-fidelity protocol are developed together.

Parallel implementations of the BB[2/7] hash-augmented tree for efficient text and state operations.

Genomic region comparison, mutation localization, tandem-repeat analysis, and drug-resistance screening.

Sub-linear context routing and dynamic state management for agentic LLMs using hash-augmented ropes.

The reference implementation of the Temporal-Logical Decay Architecture: temporal embeddings, decay-weighted retrieval, STL knowledge-validity checks, and data re-acquisition triggers for time-aware RAG. The paper is published in the IEEE IRI 2026 proceedings.

Subjective-Logic confidence algebra and a Jac/Jaseci agent walker for verifiable knowledge workflows.

Time-decayed uncertainty and evidence confidence for retrieval systems.

Air-gapped neuro-symbolic cognition over verified edge state, with Subjective Logic guardrails for LLM-driven IoT actuation. The short paper is published in the IEEE IRI 2026 proceedings.

A production-stable runtime for budgeting agent work, enforcing versioned action gates, auditing execution trees, and replaying decisions.

A bridging gateway, MCP-Lite server, and four-tier taxonomy for extending the Model Context Protocol to constrained devices.

An interactive talk and security lab covering MCP, autonomous agents, AI pipelines, attack surfaces, and defensive patterns.

The full protocol implementation: CBOR Tag 60000 annotations, constrained opinion quantization, temporal decay, cumulative fusion, MQTT and CoAP transport, security metadata, provenance chains, streaming, and batch compression.

A byte-compatible implementation for embedded, bare-metal, WASM, and ESP32 targets. Its feature tiers support full standard-library use, allocator-only environments, digest verification, or a heap-free core wire format.

Tier 1 sensors add a 4-byte opinion; Tier 2 gateways fuse and temporally decay evidence; Tier 3 services append tamper-evident provenance. Across four real IoT datasets, the accepted study reports 4.1× compression over JSON-LD for Intel telemetry and 100% IEEE 802.15.4 frame fit while retaining self-description.

CDXF provides universal binary interchange across JSON, YAML, XML, and TOML. UHC hashes directly over LZ77 streams for compressed-domain processing.

A unified explainability framework with explicit verified, quarantined, and unverified claim scopes.

A language-agnostic generator and evaluator for speech corpora with maximal phoneme coverage. An associated Show & Tell paper was accepted at Interspeech 2026.

Physics-grounded energy metrics paired with a causal, dual-rate sequence backbone for streaming sensor policies. ArcMind is an alpha research release, not a language or VLA model.

Reproducible transistor-level energy analysis for 37 signal-processing algorithms on RTX 4090 and A100 GPUs, including data and profiler traces.

A hardware-aware agent that diagnoses and repairs incompatible Python machine-learning stacks.

RGB channel multiplexing for three independent QR payloads, with neural-assisted decoding and a palette mode supporting up to nine layers. Its paper was accepted as a regular paper in ICMLA 2026 Special Session 1.

Neural steganographic data embedding in QR codes with switchable channel modes.