Primary technical disclosure
Secondary technical
Persistent Semantic Agent How the agent-resident execution substrate maintains a single persistent semantic agent identity, with cognitive state and append-only lineage preserved across device power cycles, tool lifecycle operations, and governed substrate-runtime updates under a continuity guarantee.Managed Inference Tool Registry Local data structure supporting atomic registration, deregistration, and substitution of inference tool endpoints within the agent's execution environment.Agent-to-Tool Dispatcher Dispatcher that routes inference requests from the semantic agent to registered managed endpoints based on cognitive state and policy constraints.Lineage-Derived Training Signal Training corpus constructed from downstream-outcome references with integrity-signal feedback from coupled cognitive domains.Identity Preservation Across Upgrades The semantic agent maintains continuity of identity, cognitive state, and lineage across tool lifecycle operations and substrate-runtime update operations, with the agent's identity not dependent on any specific model artifact.Cognitive State-Conditioned Dispatch Dispatcher routing conditioned on the agent's current cognitive state fields, enabling state-dependent tool selection and execution priority.Governed Tool Lifecycle Tool lifecycle operations initiated by the agent, evaluated against governance policy, and recorded in lineage with cryptographic continuity.Continuity-Proof Lineage How the agent-resident execution substrate keeps a tamper-evident lineage: an append-only record sequence under a continuity proof, optionally enforced by cryptographic chaining in which each record references its immediate predecessor.Substrate Runtime Continuity How the agent-resident execution substrate preserves the semantic agent's persistent identity, cognitive state, and append-only lineage across power cycles, application restarts, tool lifecycle operations, and substrate-runtime updates.Personal Corpus Model Training Model fine-tuned on user-authored and user-curated artifacts under governance policy, with lineage tracking from training data through derived outputs.Heterogeneous Inference Endpoints Support for multiple managed endpoint types including language, image, speech, embedding, retrieval, and personal corpus models within a single agent.Atomic Lifecycle Substitution Staged retraining with atomic promotion on success and automatic rollback on policy failure, ensuring continuity across model or tool replacements.Integrity Signal Feedback Downstream-outcome references carrying integrity-signal feedback from coupled cognitive domains into the lineage and training pipeline.Hardware-Bound Identity How the agent-resident execution substrate binds an agent's persistent identity field to a hardware security element via key derivation, with verification at subsequent substrate events and governed migration between devices.Cognitive State Append-Only Invariant How the agent-resident execution substrate preserves the semantic agent's cognitive state and lineage across lifecycle operations: an append-only lineage field, a continuity proof, and the agent as sole authority over its persistent fields.Counterparty Identity Records How the agent-resident execution substrate maintains counterparty identity records: a counterparty identifier, a scope object governing admissible interactions, an encounter history in the lineage field, a persistence designation, and attestation references verifying claimed identity.Privacy Egress-Controlled Disclosure Lineage, models, corpora, and identity records not transmitted off-device except under explicit disclosure policy evaluated against governance constraints.Federated Cross-Device Agent Identity How the agent-resident execution substrate maintains a federated agent identity across multiple devices through cross-device attestations and lineage exchange under federation policy.
Applications · general
Personal AI Agents That Survive Device Loss: One Continuous Identity and a Private Corpus Across Every Device How a personal AI agent can persist as one continuous identity with a private personal corpus across a user's devices, surviving device loss and hardware refresh, built on the Agent-Resident Execution Substrate disclosed in U.S. Provisional Application No. 64/070,239.Enterprise Agent Fleets: Stable Agent Identity and Governed Tool Access Across Model Upgrades and Infrastructure Migration How enterprise agent fleets keep stable, hardware-anchored identity and governed tool access across model upgrades and infrastructure migration, built on the Agent-Resident Execution Substrate disclosed in U.S. Provisional Application No. 64/070,239.Audit-Grade Agent Identity for Regulated Finance and Healthcare: Continuity-Proof Lineage Across the Agent Lifecycle How regulated-industry agents in finance and healthcare gain audit-grade identity and continuity-proof lineage across model upgrades and device migration, built on the Agent-Resident Execution Substrate disclosed in U.S. Provisional Application No. 64/070,239.Edge and On-Device Agents: Hardware-Bound Identity Across Heterogeneous Inference Endpoints How the Agent-Resident Execution Substrate (U.S. Provisional Application No. 64/070,239) enables edge and on-device agents to run across heterogeneous inference endpoints under one hardware-bound identity with a governed off-device disclosure boundary.Agent-to-Agent Commerce With Counterparty Identity Records and Egress-Controlled Disclosure How agent-to-agent commerce and collaboration are enabled by the counterparty identity records, egress-controlled disclosure, and compensation routing of the Agent-Resident Execution Substrate, disclosed in U.S. Provisional Application No. 64/070,239.Governed Tool Lifecycles for Managed Inference-Provider Ecosystems: A Substrate Approach to Owning, Routing, and Retiring AI Tools How a governed tool lifecycle and cognitive-state-conditioned dispatch, built on the Agent-Resident Execution Substrate (U.S. Provisional Application No. 64/070,239), turn a loose collection of model endpoints and tool providers into managed, auditable assets owned by a persistent agent.Proving Unbroken Continuity in Long-Lived Autonomous Systems Across Substrate Migration and Atomic Model Substitution How to prove an autonomous system is provably the same continuous entity across model replacement, runtime upgrade, and hardware migration, built on the Agent-Resident Execution Substrate disclosed in U.S. Provisional Application No. 64/070,239.Personal-Model Personalization: A User's Own Corpus-Internalized Model on the Agent-Resident Execution Substrate How a user's own corpus-internalized personal model, updated continuously from real authoring outcomes, is realized on the Agent-Resident Execution Substrate disclosed in U.S. Provisional Application No. 64/070,239.On-Device Agent Identity for Robots and Autonomous Vehicles: An Auditable Substrate for Embodied Physical-World Agents How the Agent-Resident Execution Substrate, disclosed in U.S. Provisional Application No. 64/070,239, provides a persistent on-device agent identity, governed inference, and an auditable lineage for robotic and autonomous-vehicle deployments.When the Agent Rents Everything It Depends On Why one drafter's assistant, assembled from rented inference endpoints, loses accumulated working judgment when a model version changes, and how the disclosed architecture keeps agent identity, outcome lineage, and a personal corpus model resident on the device instead.
Applications · specific
LangGraph Platform (LangChain) vs an agent-resident execution substrate: orchestration-graph state versus a portable, hardware-anchored agent runtime An architecture-level comparison of LangGraph Platform (LangChain) and the Agent-Resident Execution Substrate disclosed in U.S. Provisional Application No. 64/070,239, focused on where agent state and identity live.OpenAI AgentKit and the Assistants/Responses API vs agent-carried, hardware-anchored identity with governed tool lifecycle An architecture-level comparison of OpenAI AgentKit and the Assistants/Responses API against the Agent-Resident Execution Substrate disclosed in U.S. Provisional Application No. 64/070,239, focused on agent identity, lineage, and governed tool lifecycle.Microsoft Copilot Studio vs an agent-resident execution substrate: platform-hosted agent authoring versus portable, device-resident agent identity and continuity An architecture-level comparison of Microsoft Copilot Studio, a low-code agent-building platform, against an agent-resident execution substrate disclosed in U.S. Provisional Application No. 64/070,239, focused on portable agent identity and continuity versus platform-hosted authoring and orchestration.Google Vertex AI Agent Engine (managed runtime for deploying and scaling agents, with sessions/memory) vs an agent-carried, continuity-proofed identity substrate An architecture-level comparison of Google Vertex AI Agent Engine, a managed cloud agent runtime with sessions and memory, against the agent-carried continuity-proofed identity substrate disclosed in U.S. Provisional Application No. 64/070,239.AWS Bedrock AgentCore (runtime, memory, identity, and gateway services for deploying agents at scale) vs an agent-resident execution substrate: where does the agent identity actually live? How the Agent-Resident Execution Substrate of U.S. Provisional Application No. 64/070,239 differs architecturally from AWS Bedrock AgentCore: a hardware-bound, device-resident agent identity with append-only outcome lineage governing subordinate inference tools, versus managed cloud runtime, memory, identity, and gateway services.Letta (formerly MemGPT) vs an append-only cognitive-state substrate: what a memory-management framework does not provide An architecture-level comparison of Letta (formerly MemGPT), the open stateful-agent memory framework, against the append-only cognitive-state and continuity-proof lineage of the Agent-Resident Execution Substrate disclosed in U.S. Provisional Application No. 64/070,239.Cognition's Devin, an autonomous AI software-engineering agent vs a portable, continuity-proofed agent-resident runtime An architecture-level comparison of Cognition's Devin, an autonomous AI software-engineering agent, with the Agent-Resident Execution Substrate disclosed in U.S. Provisional Application No. 64/070,239.Cloudflare Agents (Durable Objects) vs an agent-resident execution substrate: portable hardware-bound identity and continuity-proof lineage An honest architecture-level comparison of Cloudflare Agents (built on Durable Objects) against the Agent-Resident Execution Substrate disclosed in U.S. Provisional Application No. 64/070,239, focused on where agent identity and lineage physically live.Ollama alternative: from local model runner to a governed agent-resident substrate How the Agent-Resident Execution Substrate (U.S. Provisional Application No. 64/070,239) differs architecturally from Ollama, contrasting a local model runner with a persistent, governed agent substrate.Apple Intelligence (on-device foundation models, Private Cloud Compute) vs a persistent agent-resident substrate: who owns identity, lineage, and the model? How the Agent-Resident Execution Substrate of U.S. Provisional Application No. 64/070,239 differs architecturally from Apple Intelligence and Private Cloud Compute along the axes of persistent agent identity, append-only inference lineage, governed model lifecycle, and a weight-internalized personal corpus model.
How-to guides
How to Build a Personal AI That Runs on Your Own Device A developer's guide to building a personal, on-device AI using an agent-resident execution substrate: a persistent local agent that governs its own inference tools, trains a personal corpus model on your work, and keeps state on-device by default.How to Build an Offline-First Personal AI That Syncs When Connected An architectural how-to for building an offline-first personal AI that runs on-device and syncs when connected, based on the agent-resident execution substrate disclosed in U.S. Provisional Application No. 64/070,239.How to Fine-Tune a Personal Model on Your Own Writing Without Sending It to the Cloud An architectural how-to for fine-tuning a personal corpus model on your own writing entirely on-device, using the lineage-derived training and privacy-invariant approach disclosed in U.S. Provisional Application No. 64/070,239.How to Give an On-Device AI Agent Tools It Can Use Safely An architectural guide to giving an on-device AI agent safe, governed tools using the agent-resident execution substrate: a managed inference tool registry with cognitive-state-conditioned dispatch and lineage-recorded outcomes.
Terminology
agent-resident execution substrate A personal computing architecture in which a semantic agent operates as the persistent runtime authority of a computing device, with inference endpoints, knowledge ingestion modules, and related computational assets maintained as governed managed components subordinate to the agent.continuity attestation A cryptographic chaining of a semantic agent's persistent identity over the sequence of substrate-runtime versions, permitting any party verifying the agent's continuity to confirm that the identity is unchanged across substrate-runtime update operations.continuity hash A cryptographic hash value maintained in a semantic agent's identity field and chained incrementally over the sequence of records appended to the lineage field.lineage-derived training corpus A training corpus assembled by selecting artifacts identified in a semantic agent's append-only lineage field, filtering the selected artifacts against the applicable corpus policy object, and applying any declared redaction or normalization rules.personal corpus model A managed inference endpoint whose model artifact parameters are trained, fine-tuned, or adapter-modified against a training corpus derived from artifacts authored, curated, or designated by the user.
