MiRuntime treats supplied reports as research inputs rather than automatically publishable truth. The index prioritizes primary sources and distinguishes foundational theory, formal proposals, prototypes, production practice, vendor implementations, MiRuntime synthesis, and forecasts.
Editorial method
- Verify time-sensitive and vendor-specific claims against primary sources.
- Attach a UTC review date, source type, status, and limitation.
- Omit unsupported metrics, product claims, anthropomorphic conclusions, and analogy presented as proof.
Program map
Research programs
Runtime infrastructure
Active execution-time control around models and agents.
Audience: platform and systems architects
Research
Runtime intelligence
Test-time compute, search, verification, and intervention.
Related: Reflective control
Research synthesis
Teleodynamic runtimes
Self-maintenance, structural adaptation, bounded goal systems, and resource-aware control.
Audience: researchers and runtime architects
Production pattern
Security and governance
Prompt injection, tools, identity, memory, authority, and evidence.
Related: Tool policy
Vendor implementation
Protocols and ecosystem
MCP, A2A, gateways, workflow systems, model serving, observability, and isolation.
Audience: integration architects
Editorial guidance
Benchmarks and reproducibility
Task success, recovery, policy, evidence, cost, latency, and replay limits.
Related: Testing guide
Status legend
MiRuntime terminology and category boundaries.
Established software-engineering control adapted to MIR.
Documented behavior of a specific product or project.
Paper, formal proposal, or prototype requiring validation.
Editorial synthesis or design position.
Explicitly speculative future scenario.
Research proposal is a specific label within the broader Research state.
Research index
33 records
Machine Intelligence Runtime
Foundational MiRuntime editorial report covering the category definition, architecture, governance, developer workflows, and research gaps.
- Topic
- Category definition
- Source type
- MiRuntime editorial report
- Reviewed
- 2026-06-20 UTC
Machine Intelligence Runtime Architectures and the Future of Artificial Intelligence
Research input spanning runtime intelligence, managed execution, protocols, security, performance, hardware, and future scenarios.
- Topic
- Future architectures
- Source type
- MiRuntime editorial report
- Reviewed
- 2026-06-20 UTC
AI Runtime Infrastructure
Research framing for an execution-time layer that observes and intervenes around agent behavior.
- Topic
- Runtime architecture
- Source type
- Research paper
- Reviewed
- 2026-06-20 UTC
Scaling LLM Test-Time Compute Optimally Can Be More Effective than Scaling Model Parameters
Primary research on allocating inference-time compute as a separate capability and systems concern.
- Topic
- Runtime intelligence
- Source type
- Research paper
- Reviewed
- 2026-06-20 UTC
Self-Consistency Improves Chain of Thought Reasoning in Language Models
Foundational inference-time sampling and aggregation method for reasoning tasks.
- Topic
- Runtime intelligence
- Source type
- Research paper
- Reviewed
- 2026-06-20 UTC
Tree of Thoughts: Deliberate Problem Solving with Large Language Models
Primary research on runtime search over intermediate reasoning states.
- Topic
- Runtime intelligence
- Source type
- Research paper
- Reviewed
- 2026-06-20 UTC
VIGIL: A Reflective Runtime for Self-Healing LLM Agents
A research prototype for reflective diagnosis and guarded repair around agent execution.
- Topic
- Recovery and reflection
- Source type
- Research paper
- Reviewed
- 2026-06-20 UTC
ATLAS-RTC: Closing the Loop on LLM Agent Output with Token-Level Runtime Control
Research on monitoring and intervening at token-generation time without changing model weights.
- Topic
- Closed-loop control
- Source type
- Research paper
- Reviewed
- 2026-06-20 UTC
Model Context Protocol Specification — 2025-11-25
Authoritative protocol requirements for connecting AI applications with tools, resources, and prompts.
- Topic
- Protocols and tools
- Source type
- Protocol specification
- Reviewed
- 2026-06-24 UTC
A2A Protocol Specification
Primary project source for communication between independently developed agent systems.
- Topic
- Agent interoperability
- Source type
- Protocol specification and reference implementation
- Reviewed
- 2026-06-24 UTC
OpenTelemetry Semantic Conventions for Generative AI Systems
Vendor-neutral semantic conventions for traces, metrics, events, and operations involving generative AI.
- Topic
- Observability
- Source type
- Official specification
- Reviewed
- 2026-06-24 UTC
Artificial Intelligence Risk Management Framework (AI RMF 1.0)
Voluntary framework for governing, mapping, measuring, and managing AI risk.
- Topic
- Risk management
- Source type
- Government framework
- Reviewed
- 2026-06-20 UTC
Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile
Companion profile describing generative-AI risk considerations and suggested actions.
- Topic
- Risk management
- Source type
- Government profile
- Reviewed
- 2026-06-20 UTC
LLM01:2025 Prompt Injection
Defines direct and indirect prompt-injection risk and mitigation considerations.
- Topic
- Runtime security
- Source type
- Security guidance
- Reviewed
- 2026-06-20 UTC
Agentic AI Threats and Mitigations
Threat-model guidance for agentic systems, including tool, identity, memory, and autonomy risks.
- Topic
- Runtime security
- Source type
- Security guidance
- Reviewed
- 2026-06-20 UTC
Amazon Bedrock AgentCore Runtime Developer Guide
Current vendor example of managed hosting and lifecycle controls for agents and tools.
- Topic
- Managed agent runtimes
- Source type
- Vendor documentation
- Reviewed
- 2026-06-20 UTC
Gemini Enterprise Agent Platform Overview
Current vendor example of a platform for deploying, governing, and observing agents.
- Topic
- Managed agent runtimes
- Source type
- Vendor documentation
- Reviewed
- 2026-06-20 UTC
Common Web Application Architectures: Clean Architecture
Official guidance describing inward dependency direction and separation of application core from infrastructure and UI.
- Topic
- Clean Architecture
- Source type
- Official architecture guidance
- Reviewed
- 2026-06-23 UTC
Cloud Design Patterns
Primary catalog for patterns including circuit breaker, CQRS, event sourcing, saga, and related distributed-system controls.
- Topic
- Distributed systems patterns
- Source type
- Official architecture guidance
- Reviewed
- 2026-06-23 UTC
Temporal Platform Documentation
Official documentation for durable workflow execution and persisted event history; cited as one compatible implementation style rather than an endorsement.
- Topic
- Durable execution
- Source type
- Official vendor documentation
- Reviewed
- 2026-06-23 UTC
MLflow Model Registry
Primary documentation for model versioning, lineage, aliases, metadata, and lifecycle management.
- Topic
- Model lifecycle and lineage
- Source type
- Official documentation
- Reviewed
- 2026-06-23 UTC
ONNX Runtime Documentation
A cross-platform inference runtime illustrating the model-serving layer below MIR.
- Topic
- Inference runtimes
- Source type
- Official documentation
- Reviewed
- 2026-06-20 UTC
vLLM Documentation
An inference and serving engine used to distinguish model scheduling from governed agent execution.
- Topic
- Inference runtimes
- Source type
- Official documentation
- Reviewed
- 2026-06-20 UTC
NVIDIA Triton Inference Server Documentation
Primary documentation for heterogeneous model serving, backends, scheduling, batching, and operational integration.
- Topic
- Inference serving
- Source type
- Vendor documentation
- Reviewed
- 2026-06-23 UTC
Triton Inference Server: Dynamic Batcher
Primary documentation for dynamic batching in an inference server.
- Topic
- Inference scheduling
- Source type
- Vendor documentation
- Reviewed
- 2026-06-20 UTC
Incomplete Nature: How Mind Emerged from Matter
Primary book-length source for Deacon’s account of constraint, morphodynamics, teleodynamics, and end-directed organization.
- Topic
- Teleodynamics
- Source type
- Foundational book
- Reviewed
- 2026-06-24 UTC
A thermodynamic basis for teleological causality
Primary contemporary articulation of teleological causality and coupled constraint in physical systems.
- Topic
- Teleodynamics
- Source type
- Peer-reviewed research paper
- Reviewed
- 2026-06-24 UTC
Autopoiesis and Cognition: The Realization of the Living
Primary source for autopoiesis and organizational self-production; runtime translations remain analogical unless explicitly formalized.
- Topic
- Autopoiesis
- Source type
- Foundational book
- Reviewed
- 2026-06-24 UTC
Autopoiesis: the organization of living systems, its characterization and a model
Original formal characterization and model of autopoietic organization in living systems.
- Topic
- Autopoiesis
- Source type
- Peer-reviewed research paper
- Reviewed
- 2026-06-24 UTC
The Embodied Mind: Cognitive Science and Human Experience
Primary source for the enactive approach to cognition and ongoing agent–environment coupling.
- Topic
- Enaction
- Source type
- Foundational book
- Reviewed
- 2026-06-24 UTC
The free-energy principle: a unified brain theory?
Foundational source for the free-energy principle; cited as a distinct computational framework rather than a synonym for teleodynamics.
- Topic
- Active inference
- Source type
- Peer-reviewed review paper
- Reviewed
- 2026-06-24 UTC
Teleodynamic Learning: a New Paradigm for Interpretable AI
Formal proposal coupling parameter adaptation, structural change, and resource constraints; production maturity and independent replication remain open.
- Topic
- Structural adaptation
- Source type
- Research preprint
- Reviewed
- 2026-06-24 UTC
Local-first software: You own your data, in spite of the cloud
Foundational local-first principles adapted here to runtime control, placement, and user ownership.
- Topic
- Local-first systems
- Source type
- Research essay
- Reviewed
- 2026-06-20 UTC
No research records match these filters.
Open research gaps
Runtime benchmarks and intervention policy
Task suites and decision policies that jointly measure success, recovery, policy, evidence, cost, and latency.
Compositional security
Trust and authority across model, tool, memory, connector, verifier, and peer-agent boundaries.
Reproducibility and stochastic replay
Evidence standards for exact, functional, partial, and unavailable replay.
Bounded goal revision
Controls that allow recovery goals without silent objective substitution.
Structural adaptation under cost
When edits outperform rerouting, retraining, or no-op after maintenance cost is included.
Self-maintenance without anthropomorphism
Measurable operating variables and falsifiable mechanisms rather than emotional language.
Metaphor versus mechanism
Formal criteria for distinguishing conceptual analogy from an implemented control system.
Long-horizon memory contamination
Detection, correction, deletion, and provenance across repeated adaptive runs.
Research data export
The public export is a reviewed editorial catalog, not a claim of exhaustive coverage. It exposes record count, topic/status values, source-review dates in UTC, and a schema version alongside stable source and citation links.
These files are conceptual MiRuntime editorial references, not a product SDK or certification profile.
These files are conceptual MiRuntime editorial references, not a product SDK or certification profile.
Source record
References
- AI Runtime Infrastructure Primary source
Christopher Cruz. arXiv. Published 2026-03; last reviewed 2026-06-20 UTC. Research paper.
Charlie Snell, Jaehoon Lee, Kelvin Xu, and Aviral Kumar. arXiv / ICLR. Published 2024-08; last reviewed 2026-06-20 UTC. Research paper.
- Model Context Protocol Specification — 2025-11-25 Primary source
Model Context Protocol project. Model Context Protocol. Published 2025-11-25; last reviewed 2026-06-24 UTC. Protocol specification.
National Institute of Standards and Technology. NIST. Published 2023-01-26; last reviewed 2026-06-20 UTC. Government framework.