See the whole stack.

KVM-based microVMs for isolating lightweight compute workloads.

Stateful agent orchestration with explicit control flow, checkpoints and human review.

Programmatic cloud sandboxes for executing agent-generated code.

Temporal knowledge graphs for agent context that changes over time.

Agent loops, model abstractions and integrations built on LangGraph.

Traces, evaluations and prompt management for agent applications.

Local C/C++ inference and an HTTP server across multiple hardware backends.

Persistent, retrievable memories for users, sessions and agent applications.

Browser automation and testing across Chromium, Firefox and WebKit.

Configurable evaluations and red-team tests for models and agent applications.

A typed Python agent framework with validated outputs and composable tools.

Durable execution for workflows that must recover across failures and long waits.

Model inference and serving with batching and distributed execution options.

Listed in Inference Infrastructure · scale-out.

Listed in Inference Infrastructure · scale-out.

Listed in Agent Connectivity & State · agent connectivity.

Listed in Agent Harness · agent builders & interfaces.

Listed in Agent Harness · agent builders & interfaces.

Listed in Agent Connectivity & State · drafts & workstreams.

Listed in Agent Harness · agent rl & self-improvement.

Published community specification for reusable agent skills; not an ISO, IETF, or W3C standard.

Sandboxed, stateful agent execution runtime; separate project required by AX.

General multimodal agent stack with CLI and web UI; shares a repository with UI-TARS Desktop but is a distinct deliverable. Classified with agent builders and interfaces.

Listed in Agent Connectivity & State · agent connectivity.

Published community format for agent instruction files; not an ISO, IETF, or W3C standard.

Listed in Agent Harness · agent rl & self-improvement.

Broader inference orchestration platform with a KV-cache offloading component.

Listed in Agent Connectivity & State · drafts & workstreams.

Listed in Agent Connectivity & State · drafts & workstreams.

Published Agent Network Protocol documents; individual profiles have their own status and maturity.

Listed in Agent Harness · agent rl & self-improvement.

Listed in Agent Harness · harness & turn management.

Listed in Agent Harness · harness & turn management.

Pre-1.0, declarative agent workload orchestrator for clusters; relies on Agent Substrate for sandboxed execution.

Listed in Open Models · training & fine-tuning.

Listed in Inference Infrastructure · scale-out.

Browser agent and automation library; its Browser Actor uses the Chrome DevTools Protocol directly.

Web-agent research and evaluation environment; uses Playwright for browser setup.

Listed in Agent Execution Environment · sandboxes & runtimes.

Public CANN community components; the name does not assert that every CANN component is open source.

Listed in Agent Connectivity & State · drafts & workstreams.

Listed in Identity & Access · identity & access.

CloudEvents on-the-wire specversion is 1.0; the published specification has a 1.0.2 patch release.

Listed in Agent Harness · harness & turn management.

Listed in Agent Harness · memory & context.

Container lifecycle manager that delegates launching containers to lower-level runtimes.

Listed in Agent Harness · agent builders & interfaces.

Listed in Agent Harness · harness & turn management.

Published cache-coherent interconnect specification for processors, memory expansion, and accelerators.

Listed in Trace & Evidence · trace & evidence.

Separate Python framework for building durable AI agents on the general-purpose Dapr runtime.

Listed in Open Models · training & fine-tuning.

Listed in Open Models · training & fine-tuning.

Listed in Trace & Evidence · trace & evidence.

Listed in Open Models · model families.

Listed in Agent Harness · harness & turn management.

Listed in Open Models · training & fine-tuning.

Listed in Agent Harness · harness & turn management.

General-purpose decentralized HTTP transport included as agent connectivity infrastructure, not an agent-only protocol.

Public source; current license is modified Apache 2.0 with additional conditions.

Listed in Inference Infrastructure · management & trust.

Listed in Inference Infrastructure · management & trust.

Listed in Inference Infrastructure · management & trust.

Listed in Agent Connectivity & State · internet.

Listed in Agent Connectivity & State · internet.

Listed in Agent Connectivity & State · internet.

Listed in Agent Execution Environment · durable & dataflow execution.

Listed in Agent Connectivity & State · drafts & workstreams.

Listed in Agent Harness · agent rl & self-improvement.

Listed in Inference Infrastructure · inference engines & serving.

Listed in Trace & Evidence · trace & evidence.

Listed in Inference Infrastructure · scale-out.

Combined deployment pattern supported by separate EVPN and VXLAN RFCs, not the name of one standalone RFC.

Umbrella AI system software stack; components are maintained in separate public repositories.

Inference KV-cache manager; not persistent agent memory.

Listed in Open Models · model families.

Documented model file format used by GGML-based inference engines; not a separate software project.

Open model family published by Z.ai; Z.ai is the publisher, not part of the model-family name.

Listed in Identity & Access · identity & access.

Listed in Inference Infrastructure · management & trust.

Code-first framework for building, evaluating, and deploying agents; it is broader than a standalone harness.

Listed in Agent Harness · harness & turn management.

Listed in Agent Execution Environment · sandboxes & runtimes.

Listed in Agent Harness · harness & turn management.

Listed in Agent Connectivity & State · internet.

Listed in Agent Connectivity & State · internet.

Reader-facing label for the open-source huggingface_hub CLI and Python client, not a claim that the hosted Hub service is open source.

Listed in Inference Infrastructure · scale-up.

Listed in Agent Connectivity & State · drafts & workstreams.

Listed in Trace & Evidence · trace & evidence.

Listed in Inference Infrastructure · scale-out.

Listed in Agent Connectivity & State · internet.

Listed in Inference Infrastructure · management & trust.

Listed in Agent Harness · agent rl & self-improvement.

Official project name for the separately released openJiuwen agent-memory repository.

Listed in Intent Contract · intent contract.

Two related IETF specifications, RFC 7519 and RFC 7515; catalog source points to JWT and the second RFC is linked below.

Listed in Agent Execution Environment · sandboxes & runtimes.

Listed in Identity & Access · identity & access.

Listed in Open Models · model families.

Listed in Inference Infrastructure · inference engines & serving.

Listed in Inference Infrastructure · gpu cluster orchestration.

Listed in Inference Infrastructure · gpu cluster orchestration.

Listed in Open Models · model families.

Listed in Agent Harness · memory & context.

Self-hosted AI application and interface with a no-code Agent Builder; not a general-purpose agent harness.

Listed in Agent Execution Environment · operating systems & endpoints.

Listed in Open Models · model families.

Listed in Open Models · training & fine-tuning.

Listed in Inference Infrastructure · inference engines & serving.

Inference KV-cache storage and reuse; not persistent agent memory.

Listed in Agent Harness · agent builders & interfaces.

Listed in Agent Harness · harness & turn management.

General-purpose federated communication standard; shown here as connectivity infrastructure, not as an agent-specific protocol.

Listed in Agent Connectivity & State · agent connectivity.

Listed in Open Models · training & fine-tuning.

Listed in Agent Harness · memory & context.

Listed in Agent Harness · memory & context.

Agent framework with workflow orchestration and harness capabilities; distinct from its predecessor AutoGen.

Listed in Inference Infrastructure · inference engines & serving.

Listed in Open Models · training & fine-tuning.

Listed in Open Models · model families.

Listed in Open Models · model families.

Listed in Open Models · model families.

Listed in Inference Infrastructure · inference engines & serving.

Open-source container engine components; broader than a sandbox or low-level runtime.

Open-source ModelScope library with model hub access; not a claim that the hosted platform is open source.

Listed in Inference Infrastructure · kv cache, storage & data movement.

Transfer Engine component of the Mooncake repository; distinct from Mooncake Store and its KV cache use.

Listed in Open Models · model families.

General-purpose messaging infrastructure used by agents; not an agent-specific protocol.

Listed in Open Models · model families.

Listed in Agent Connectivity & State · drafts & workstreams.

Listed in Inference Infrastructure · kv cache, storage & data movement.

The operator source is open; deployed NVIDIA drivers and CUDA components may have separate licenses.

Listed in Inference Infrastructure · storage.

The current NVM Express Management Interface specification is revision 2.2.

Listed in Inference Infrastructure · storage.

Listed in Inference Infrastructure · storage.

Listed in Identity & Access · identity & access.

Listed in Trace & Evidence · trace & evidence.

Listed in Agent Execution Environment · execution specs.

Listed in Agent Execution Environment · execution specs.

Listed in Agent Execution Environment · execution specs.

Listed in Inference Infrastructure · scale-up.

Listed in Inference Infrastructure · management & trust.

Listed in Agent Harness · agent builders & interfaces.

Listed in Agent Harness · harness & turn management.

Listed in Identity & Access · identity & access.

Listed in Open Models · model families.

Listed in Agent Execution Environment · operating systems & endpoints.

Listed in Open Models · formats.

Listed in Identity & Access · identity & access.

Public source; releases after v0.6.5 have additional branding conditions.

Listed in Agent Connectivity & State · agent connectivity.

Listed in Intent Contract · intent contract.

Listed in Identity & Access · identity & access.

Listed in Agent Harness · harness & turn management.

Listed in Inference Infrastructure · management & trust.

Listed in Agent Execution Environment · operating systems & endpoints.

In-toto predicate for AI code-generation provenance: per-range human/AI attribution, prompt hash binding, a signed embedded acceptance contract, and cryptographic human sign-offs. Registration as a vetted in-toto predicate type is proposed and under maintainer review (in-toto/attestation#604); shown here as a project pending that outcome.

Listed in Identity & Access · identity & access.

Listed in Agent Harness · harness & turn management.

Listed in Agent Execution Environment · operating systems & endpoints.

Listed in Agent Harness · harness & turn management.

Listed in Open Models · model families.

Listed in Trace & Evidence · trace & evidence.

Listed in Agent Execution Environment · operating systems & endpoints.

Listed in Inference Infrastructure · drafts & workstreams.

Listed in Inference Infrastructure · scale-up.

Listed in Open Models · training & fine-tuning.

Listed in Agent Harness · agent rl & self-improvement.

Listed in Trace & Evidence · trace & evidence.

Listed in Agent Harness · agent rl & self-improvement.

Listed in Agent Harness · harness & turn management.

MCP server exposing Playwright browser automation to agents; a separate project built on Playwright.

Listed in Agent Execution Environment · execution specs.

WebAssembly-based Python distribution for browsers and Node.js; not a security sandbox by itself.

Listed in Inference Infrastructure · ai compute & kernel programming.

Listed in Agent Connectivity & State · internet.

Listed in Open Models · model families.

Listed in Trace & Evidence · trace & evidence.

Listed in Agent Harness · agent rl & self-improvement.

Listed in Trace & Evidence · trace & evidence.

Listed in Inference Infrastructure · storage.

Listed in Inference Infrastructure · management & trust.

Listed in Agent Harness · harness & turn management.

Open-source Matrix chat client built with Makepad; a connectivity application, not an agent-specific protocol.

Listed in Inference Infrastructure · scale-out.

AMD open software stack; licenses vary by component.

Listed in Agent Connectivity & State · drafts & workstreams.

Display label for the safetensors tensor serialization format and its open-source library; the official repository and package name is lowercase plural “safetensors”.

Listed in Identity & Access · identity & access.

Listed in Trace & Evidence · trace & evidence.

Listed in Inference Infrastructure · inference engines & serving.

Listed in Trace & Evidence · trace & evidence.

Listed in Agent Harness · agent rl & self-improvement.

Listed in Trace & Evidence · trace & evidence.

Listed in Inference Infrastructure · gpu cluster orchestration.

Listed in Inference Infrastructure · kv cache, storage & data movement.

Listed in Trace & Evidence · trace & evidence.

Listed in Identity & Access · identity & access.

Listed in Identity & Access · identity & access.

Listed in Agent Connectivity & State · internet.

The initial contributed Scale-Up Ethernet specification; successor effort is the SUE-T workstream.

Listed in Inference Infrastructure · drafts & workstreams.

Listed in Agent Connectivity & State · internet.

Listed in Inference Infrastructure · ai compute & kernel programming.

Listed in Agent Connectivity & State · internet.

Listed in Open Models · training & fine-tuning.

Listed in Open Models · training & fine-tuning.

Listed in Inference Infrastructure · ai compute & kernel programming.

Listed in Open Models · training & fine-tuning.

Listed in Agent Connectivity & State · drafts & workstreams.

AI/IoT device development framework with agent-capable applications; not an operating system.

Listed in Inference Infrastructure · scale-up.

Listed in Inference Infrastructure · scale-up.

Listed in Inference Infrastructure · scale-up.

Listed in Inference Infrastructure · scale-up.

Listed in Inference Infrastructure · scale-up.

Listed in Inference Infrastructure · scale-up.

Listed in Inference Infrastructure · scale-up.

Unified Cache Manager for inference KV cache; not persistent agent memory.

Listed in Inference Infrastructure · kv cache, storage & data movement.

Desktop GUI agent application; shares a repository with Agent TARS but is a distinct deliverable.

Listed in Inference Infrastructure · scale-out.

Listed in Inference Infrastructure · scale-up.

Listed in Open Models · training & fine-tuning.

Public source under BUSL 1.1 for current releases; not an unrestricted open-source license.

Listed in Open Models · training & fine-tuning.

Listed in Inference Infrastructure · gpu cluster orchestration.

Listed in Identity & Access · identity & access.

Listed in Trace & Evidence · trace & evidence.

Listed in Identity & Access · identity & access.

Listed in Agent Execution Environment · execution specs.

Listed in Agent Execution Environment · sandboxes & runtimes.

Listed in Inference Infrastructure · management & trust.

General-purpose distributed data and messaging middleware used by agents; not an agent-specific protocol.

Listed in Agent Harness · harness & turn management.