{"name":"Anomra Atlas","description":"Six dimensions shape every multi-agent system: orchestration, memory, routing, governance, observability, and human-in-the-loop. This atlas maps how the leading frameworks handle each, with orchestration as the lead lens.","url":"https://www.anomra.com/atlas","dimensions":[{"id":"orchestration","name":"Orchestration","summary":"How agents are coordinated","description":"How work is decomposed and coordinated across agents: supervisor graphs, sequential chains, parallel fan-out, loops, handoff networks, and role hierarchies."},{"id":"memory","name":"Memory","summary":"What persists across steps","description":"What persists across steps and sessions: run state, checkpoints, short and long term stores, and recalled context."},{"id":"routing","name":"Routing","summary":"How control reaches an agent","description":"How control and data reach the right agent or tool: conditional edges, handoffs, manager selection, and delegation."},{"id":"governance","name":"Governance","summary":"Guardrails, policy, approval","description":"The guardrails, policies, validation, and approval gates that keep agents inside their bounds."},{"id":"observability","name":"Observability","summary":"Tracing and telemetry","description":"The tracing, telemetry, and evaluation that make an agent run inspectable and replayable."},{"id":"hitl","name":"Human in the loop","summary":"Where a person steps in","description":"Where a person reviews, approves, edits, or interrupts the run before it continues."}],"frameworks":[{"id":"langgraph","name":"LangGraph","pattern":"Stateful supervisor graph","essence":"Agents are nodes in a directed state graph; a supervisor node routes to workers and loops until the goal state is reached.","docUrl":"https://langchain-ai.github.io/langgraph/","runtimeNote":null,"byDimension":{"orchestration":"A StateGraph of nodes and edges. A supervisor node routes to worker nodes and the graph loops until it reaches END.","memory":"Checkpointers persist graph state per thread, enabling pause, resume, and time-travel debugging.","routing":"Conditional edges choose the next node from the current state.","governance":"Validation lives in nodes; interrupts gate sensitive steps before they run.","observability":"LangSmith tracing captures every node transition and state diff.","hitl":"interrupt and interrupt_before pause the graph for human review, then resume from the checkpoint."}},{"id":"crewai","name":"CrewAI","pattern":"Role-based crew","essence":"Agents have a role, goal, and backstory; a process runs them sequentially or hands them to a manager agent that delegates and collects.","docUrl":"https://docs.crewai.com/","runtimeNote":null,"byDimension":{"orchestration":"Role-based agents run under a Process: sequential task order, or hierarchical with a manager agent delegating to the crew.","memory":"Built-in short-term, long-term, and entity memory; long-term memory persists across runs in a local store.","routing":"The manager (hierarchical) or task sequence (sequential) routes work; agents can delegate to peers.","governance":"Each task declares expected_output and can attach guardrails that validate results.","observability":"Verbose logging and event listeners; integrates with tracing tools such as AgentOps.","hitl":"Setting human_input on a task pauses execution for operator feedback."}},{"id":"autogen","name":"AutoGen","pattern":"Conversational group chat","essence":"Agents collaborate in a shared conversation; a manager or selector policy picks who speaks next each turn.","docUrl":"https://microsoft.github.io/autogen/","runtimeNote":null,"byDimension":{"orchestration":"Agents converse in a group chat; a manager selects the next speaker. v0.4 adds an actor-model runtime and reusable Teams (RoundRobin, Selector).","memory":"Conversation history is the working memory; external memory stores are opt-in.","routing":"RoundRobinGroupChat or SelectorGroupChat chooses the next agent; handoffs are supported.","governance":"Termination conditions and allowed-speaker rules bound the chat; code runs in sandboxes.","observability":"v0.4 emits OpenTelemetry traces alongside message logging.","hitl":"A UserProxyAgent can require human input or approval mid-conversation."}},{"id":"openai-agents","name":"OpenAI Agents","pattern":"Agents with handoffs","essence":"Small agents with instructions and tools pass control to one another through explicit handoffs, fronted by guardrails.","docUrl":"https://openai.github.io/openai-agents-python/","runtimeNote":null,"byDimension":{"orchestration":"Lightweight agents with instructions and tools; a triage agent passes control through explicit handoffs.","memory":"Sessions automatically persist conversation history across runs.","routing":"Handoffs are modeled as tools; a triage agent routes to specialists.","governance":"Input and output guardrails validate and can halt a run; tool schemas are typed.","observability":"Built-in tracing visualizes runs, handoffs, and tool calls.","hitl":"Tool calls can require human approval before they execute."}},{"id":"anthropic","name":"Claude / Anthropic","pattern":"Orchestrator and workers","essence":"Composable workflow patterns plus autonomous agents: an orchestrator dynamically splits a task into worker calls, then synthesizes the result.","docUrl":"https://www.anthropic.com/engineering/building-effective-agents","runtimeNote":null,"byDimension":{"orchestration":"Composable workflow patterns - prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer - plus autonomous agents for open-ended tasks.","memory":"Stateless by default; context is passed explicitly, with files and external memory added when needed.","routing":"A router classifies input to a path; an orchestrator decomposes tasks to workers at run time.","governance":"Guardrails through tool design and evaluator-optimizer loops; the Agent SDK scopes permissions and subagents.","observability":"Tracing through the API and Agent SDK; evaluator steps score intermediate outputs.","hitl":"Patterns favor explicit human checkpoints; the Agent SDK supports permissioned, reviewable tool use."}},{"id":"google-adk","name":"Gemini / Google ADK","pattern":"Composable agent hierarchy","essence":"Compose LlmAgents with workflow primitives - Sequential, Parallel, and Loop agents - into multi-agent hierarchies.","docUrl":"https://google.github.io/adk-docs/","runtimeNote":null,"byDimension":{"orchestration":"Compose agents with workflow primitives - SequentialAgent, ParallelAgent, LoopAgent - around LlmAgents to form multi-agent hierarchies.","memory":"Session state plus pluggable Memory services; Vertex AI Agent Engine offers managed sessions and memory.","routing":"LLM-driven transfer between sub-agents, or deterministic order through workflow agents.","governance":"Before and after callbacks and plugins gate model, tool, and agent steps; Vertex adds safety controls.","observability":"Built-in evaluation and tracing that integrate with Cloud tracing.","hitl":"Human approval through tools and callbacks; long-running operations can await input."}},{"id":"nemo-agent-toolkit","name":"NVIDIA NeMo Agent Toolkit","pattern":"Framework-agnostic meta-orchestrator","essence":"A framework-agnostic toolkit that sits over agents built in other frameworks: a YAML-defined workflow connects, profiles, and optimizes them, adding routing, pluggable memory, observability, and human approval.","docUrl":"https://developer.nvidia.com/nemo-agent-toolkit","runtimeNote":"Serving: the models a NeMo workflow calls are typically deployed behind NVIDIA NIM inference microservices - containerized, OpenAI-compatible inference endpoints, part of NVIDIA AI Enterprise - and NeMo Agent Toolkit can weave NIM microservices into a workflow.","byDimension":{"orchestration":"A YAML config defines a workflow that composes functions, LLMs, and agents (ReAct, Reasoning, ReWOO, Router, Tool Calling), run with `nat run`. It is a framework-agnostic meta-orchestrator over agents built in other frameworks, and can expose them over MCP or A2A.","memory":"Memory is pluggable rather than built in: integrations such as Zep, Mem0, and Memmachine attach to any agent or tool through the config.","routing":"A Router agent and node-level routing send each request to the right tool or sub-agent; distributed routing runs over MCP servers and the Agent-to-Agent (A2A) protocol.","governance":"There is no dedicated governance product; the documented controls are human-in-the-loop approval gates plus middleware such as defense and red-teaming added to the workflow.","observability":"Its core purpose: profile and trace a workflow down to individual tokens, streaming OpenTelemetry traces with built-in exporters for Phoenix, Weave, Langfuse, and LangSmith.","hitl":"It can add human-in-the-loop interaction to any tool or function to collect input or approval mid-run, without coding it into the agent itself."}},{"id":"business-os","name":"Business OS pattern","pattern":"Governed control plane","essence":"A supervisor pulls tasks from a work queue and dispatches them to worker agents that run business operations behind policy gates and an audit ledger.","docUrl":null,"runtimeNote":null,"byDimension":{"orchestration":"A supervisor pulls tasks from a work queue and dispatches them to worker agents that carry out business operations.","memory":"Durable state lives in the queue, task records, and an append-only audit ledger.","routing":"The supervisor assigns by capability; every result flows through a policy gate.","governance":"Policy and compliance gates validate each action and require approval for risky steps - the defining trait of the pattern.","observability":"Audit ledgers and telemetry make every decision traceable and replayable.","hitl":"Operators approve, reject, or amend at explicit gates before actions commit."}}],"chooseByControlSubstrate":[{"substrate":"A graph over shared state","frameworks":"LangGraph","when":"you need durable, resumable, inspectable runs with human checkpoints"},{"substrate":"A conversation","frameworks":"AutoGen","when":"collaboration is open-ended and you want emergent back-and-forth"},{"substrate":"Handoffs between specialists","frameworks":"OpenAI Agents, AutoGen Swarm","when":"a few agents pass control and you want minimal machinery"},{"substrate":"A role hierarchy","frameworks":"CrewAI","when":"the work maps to an org chart of roles and tasks"},{"substrate":"A composable workflow","frameworks":"Claude / Anthropic, Google ADK","when":"you mix deterministic steps with model decisions and add complexity only as needed"},{"substrate":"A wrapper over other frameworks","frameworks":"NVIDIA NeMo Agent Toolkit","when":"you want to connect, profile, and optimize agents built in different frameworks without rewriting them"},{"substrate":"A governed control plane","frameworks":"Business OS","when":"operations must be audited and approved - it wraps any of the above"}],"decisionGuide":[{"need":"Durable, resumable runs you can pause, inspect, and hand to a human","id":"langgraph","framework":"LangGraph"},{"need":"A team of role-based agents working a task list, fast","id":"crewai","framework":"CrewAI"},{"need":"Open-ended collaboration where agents talk it through","id":"autogen","framework":"AutoGen"},{"need":"A few specialists that hand off, with tracing built in","id":"openai-agents","framework":"OpenAI Agents"},{"need":"Start with the simplest workflow and add agents only when needed","id":"anthropic","framework":"Claude / Anthropic"},{"need":"Deterministic and model-driven control on Google Cloud, with A2A","id":"google-adk","framework":"Gemini / Google ADK"},{"need":"Connect, profile, and optimize agents from different frameworks without rewriting them","id":"nemo-agent-toolkit","framework":"NVIDIA NeMo Agent Toolkit"},{"need":"Operations that must be governed, audited, and approved","id":"business-os","framework":"Business OS pattern"}],"maps":[{"id":"agent-frameworks","title":"Agent frameworks","thesis":"Six dimensions shape every multi-agent system: orchestration, memory, routing, governance, observability, and human-in-the-loop. This atlas maps how the leading frameworks handle each, with orchestration as the lead lens.","dimensions":[{"id":"orchestration","name":"Orchestration","summary":"How agents are coordinated","description":"How work is decomposed and coordinated across agents: supervisor graphs, sequential chains, parallel fan-out, loops, handoff networks, and role hierarchies."},{"id":"memory","name":"Memory","summary":"What persists across steps","description":"What persists across steps and sessions: run state, checkpoints, short and long term stores, and recalled context."},{"id":"routing","name":"Routing","summary":"How control reaches an agent","description":"How control and data reach the right agent or tool: conditional edges, handoffs, manager selection, and delegation."},{"id":"governance","name":"Governance","summary":"Guardrails, policy, approval","description":"The guardrails, policies, validation, and approval gates that keep agents inside their bounds."},{"id":"observability","name":"Observability","summary":"Tracing and telemetry","description":"The tracing, telemetry, and evaluation that make an agent run inspectable and replayable."},{"id":"hitl","name":"Human in the loop","summary":"Where a person steps in","description":"Where a person reviews, approves, edits, or interrupts the run before it continues."}],"entries":[{"id":"langgraph","name":"LangGraph","pattern":"Stateful supervisor graph","essence":"Agents are nodes in a directed state graph; a supervisor node routes to workers and loops until the goal state is reached.","docUrl":"https://langchain-ai.github.io/langgraph/","runtimeNote":null,"byDimension":{"orchestration":"A StateGraph of nodes and edges. A supervisor node routes to worker nodes and the graph loops until it reaches END.","memory":"Checkpointers persist graph state per thread, enabling pause, resume, and time-travel debugging.","routing":"Conditional edges choose the next node from the current state.","governance":"Validation lives in nodes; interrupts gate sensitive steps before they run.","observability":"LangSmith tracing captures every node transition and state diff.","hitl":"interrupt and interrupt_before pause the graph for human review, then resume from the checkpoint."}},{"id":"crewai","name":"CrewAI","pattern":"Role-based crew","essence":"Agents have a role, goal, and backstory; a process runs them sequentially or hands them to a manager agent that delegates and collects.","docUrl":"https://docs.crewai.com/","runtimeNote":null,"byDimension":{"orchestration":"Role-based agents run under a Process: sequential task order, or hierarchical with a manager agent delegating to the crew.","memory":"Built-in short-term, long-term, and entity memory; long-term memory persists across runs in a local store.","routing":"The manager (hierarchical) or task sequence (sequential) routes work; agents can delegate to peers.","governance":"Each task declares expected_output and can attach guardrails that validate results.","observability":"Verbose logging and event listeners; integrates with tracing tools such as AgentOps.","hitl":"Setting human_input on a task pauses execution for operator feedback."}},{"id":"autogen","name":"AutoGen","pattern":"Conversational group chat","essence":"Agents collaborate in a shared conversation; a manager or selector policy picks who speaks next each turn.","docUrl":"https://microsoft.github.io/autogen/","runtimeNote":null,"byDimension":{"orchestration":"Agents converse in a group chat; a manager selects the next speaker. v0.4 adds an actor-model runtime and reusable Teams (RoundRobin, Selector).","memory":"Conversation history is the working memory; external memory stores are opt-in.","routing":"RoundRobinGroupChat or SelectorGroupChat chooses the next agent; handoffs are supported.","governance":"Termination conditions and allowed-speaker rules bound the chat; code runs in sandboxes.","observability":"v0.4 emits OpenTelemetry traces alongside message logging.","hitl":"A UserProxyAgent can require human input or approval mid-conversation."}},{"id":"openai-agents","name":"OpenAI Agents","pattern":"Agents with handoffs","essence":"Small agents with instructions and tools pass control to one another through explicit handoffs, fronted by guardrails.","docUrl":"https://openai.github.io/openai-agents-python/","runtimeNote":null,"byDimension":{"orchestration":"Lightweight agents with instructions and tools; a triage agent passes control through explicit handoffs.","memory":"Sessions automatically persist conversation history across runs.","routing":"Handoffs are modeled as tools; a triage agent routes to specialists.","governance":"Input and output guardrails validate and can halt a run; tool schemas are typed.","observability":"Built-in tracing visualizes runs, handoffs, and tool calls.","hitl":"Tool calls can require human approval before they execute."}},{"id":"anthropic","name":"Claude / Anthropic","pattern":"Orchestrator and workers","essence":"Composable workflow patterns plus autonomous agents: an orchestrator dynamically splits a task into worker calls, then synthesizes the result.","docUrl":"https://www.anthropic.com/engineering/building-effective-agents","runtimeNote":null,"byDimension":{"orchestration":"Composable workflow patterns - prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer - plus autonomous agents for open-ended tasks.","memory":"Stateless by default; context is passed explicitly, with files and external memory added when needed.","routing":"A router classifies input to a path; an orchestrator decomposes tasks to workers at run time.","governance":"Guardrails through tool design and evaluator-optimizer loops; the Agent SDK scopes permissions and subagents.","observability":"Tracing through the API and Agent SDK; evaluator steps score intermediate outputs.","hitl":"Patterns favor explicit human checkpoints; the Agent SDK supports permissioned, reviewable tool use."}},{"id":"google-adk","name":"Gemini / Google ADK","pattern":"Composable agent hierarchy","essence":"Compose LlmAgents with workflow primitives - Sequential, Parallel, and Loop agents - into multi-agent hierarchies.","docUrl":"https://google.github.io/adk-docs/","runtimeNote":null,"byDimension":{"orchestration":"Compose agents with workflow primitives - SequentialAgent, ParallelAgent, LoopAgent - around LlmAgents to form multi-agent hierarchies.","memory":"Session state plus pluggable Memory services; Vertex AI Agent Engine offers managed sessions and memory.","routing":"LLM-driven transfer between sub-agents, or deterministic order through workflow agents.","governance":"Before and after callbacks and plugins gate model, tool, and agent steps; Vertex adds safety controls.","observability":"Built-in evaluation and tracing that integrate with Cloud tracing.","hitl":"Human approval through tools and callbacks; long-running operations can await input."}},{"id":"nemo-agent-toolkit","name":"NVIDIA NeMo Agent Toolkit","pattern":"Framework-agnostic meta-orchestrator","essence":"A framework-agnostic toolkit that sits over agents built in other frameworks: a YAML-defined workflow connects, profiles, and optimizes them, adding routing, pluggable memory, observability, and human approval.","docUrl":"https://developer.nvidia.com/nemo-agent-toolkit","runtimeNote":"Serving: the models a NeMo workflow calls are typically deployed behind NVIDIA NIM inference microservices - containerized, OpenAI-compatible inference endpoints, part of NVIDIA AI Enterprise - and NeMo Agent Toolkit can weave NIM microservices into a workflow.","byDimension":{"orchestration":"A YAML config defines a workflow that composes functions, LLMs, and agents (ReAct, Reasoning, ReWOO, Router, Tool Calling), run with `nat run`. It is a framework-agnostic meta-orchestrator over agents built in other frameworks, and can expose them over MCP or A2A.","memory":"Memory is pluggable rather than built in: integrations such as Zep, Mem0, and Memmachine attach to any agent or tool through the config.","routing":"A Router agent and node-level routing send each request to the right tool or sub-agent; distributed routing runs over MCP servers and the Agent-to-Agent (A2A) protocol.","governance":"There is no dedicated governance product; the documented controls are human-in-the-loop approval gates plus middleware such as defense and red-teaming added to the workflow.","observability":"Its core purpose: profile and trace a workflow down to individual tokens, streaming OpenTelemetry traces with built-in exporters for Phoenix, Weave, Langfuse, and LangSmith.","hitl":"It can add human-in-the-loop interaction to any tool or function to collect input or approval mid-run, without coding it into the agent itself."}},{"id":"business-os","name":"Business OS pattern","pattern":"Governed control plane","essence":"A supervisor pulls tasks from a work queue and dispatches them to worker agents that run business operations behind policy gates and an audit ledger.","docUrl":null,"runtimeNote":null,"byDimension":{"orchestration":"A supervisor pulls tasks from a work queue and dispatches them to worker agents that carry out business operations.","memory":"Durable state lives in the queue, task records, and an append-only audit ledger.","routing":"The supervisor assigns by capability; every result flows through a policy gate.","governance":"Policy and compliance gates validate each action and require approval for risky steps - the defining trait of the pattern.","observability":"Audit ledgers and telemetry make every decision traceable and replayable.","hitl":"Operators approve, reject, or amend at explicit gates before actions commit."}}],"decisionGuide":[{"need":"Durable, resumable runs you can pause, inspect, and hand to a human","id":"langgraph","entry":"LangGraph"},{"need":"A team of role-based agents working a task list, fast","id":"crewai","entry":"CrewAI"},{"need":"Open-ended collaboration where agents talk it through","id":"autogen","entry":"AutoGen"},{"need":"A few specialists that hand off, with tracing built in","id":"openai-agents","entry":"OpenAI Agents"},{"need":"Start with the simplest workflow and add agents only when needed","id":"anthropic","entry":"Claude / Anthropic"},{"need":"Deterministic and model-driven control on Google Cloud, with A2A","id":"google-adk","entry":"Gemini / Google ADK"},{"need":"Connect, profile, and optimize agents from different frameworks without rewriting them","id":"nemo-agent-toolkit","entry":"NVIDIA NeMo Agent Toolkit"},{"need":"Operations that must be governed, audited, and approved","id":"business-os","entry":"Business OS pattern"}]},{"id":"physical-ai","title":"Physical AI","thesis":"Physical AI asks a different set of questions than agent software. These stacks are mapped across six dimensions - perception, world model, simulation and digital twins, policy and control, sim-to-real, and safety - with the world model as the lead lens. Several are research-stage; where a stack does not target a dimension, the map says so.","dimensions":[{"id":"perception","name":"Perception","summary":"Sensing the world","description":"How the system takes in the physical world: cameras, sensors, video, or multimodal capture turned into a usable representation."},{"id":"worldModel","name":"World model","summary":"How the world evolves","description":"A learned model that predicts how a scene evolves and how actions change it, used to imagine or roll out future states."},{"id":"simulation","name":"Simulation","summary":"Simulation and digital twins","description":"Physically based virtual environments and digital twins where behaviour is generated, rendered, and tested before the real world."},{"id":"policyControl","name":"Policy & control","summary":"Learning and choosing actions","description":"How an embodied agent's actions are learned or produced: reinforcement and imitation learning, or planning toward a goal."},{"id":"simToReal","name":"Sim-to-real","summary":"Crossing into the real world","description":"Transferring behaviour learned in simulation or a world model onto real hardware, closing the gap to physical reality."},{"id":"safety","name":"Safety","summary":"Safety and guardrails","description":"Validating behaviour and constraining outputs: simulation-based testing, content guardrails, and staged, reviewed release."}],"entries":[{"id":"nvidia-physical-ai","name":"NVIDIA","pattern":"Full stack: simulate, model, train, deploy","essence":"An integrated, full-stack approach to physical AI: Omniverse and Isaac Sim build OpenUSD simulations and digital twins, Cosmos world foundation models predict and generate physical worlds, and Isaac Lab trains robot policies that transfer to real hardware.","docUrl":"https://developer.nvidia.com/isaac","runtimeNote":"Serving: Cosmos world foundation models are available as NVIDIA NIM inference microservices (Cosmos Reason ships as a NIM) for production deployment across cloud, data centre, and workstation.","byDimension":{"perception":"Isaac Sim simulates cameras and sensors and generates annotated synthetic data (RGB, segmentation, bounding boxes) to train perception stacks; Cosmos Reason adds a vision-language model that reasons over scenes.","worldModel":"Cosmos world foundation models understand and simulate physical environments and predict future world states from text, image, video, and action inputs.","simulation":"Omniverse and Isaac Sim build physically based simulations and digital twins on the OpenUSD scene standard, with PhysX / Newton physics and RTX rendering.","policyControl":"Isaac Lab is a GPU-accelerated framework for training robot policies at scale with reinforcement and imitation learning across humanoids, arms, and mobile robots.","simToReal":"Isaac Sim ingests CAD, URDF, and real-world captures into USD and supports software- and hardware-in-the-loop, while Isaac Lab trains policies in sim to bridge the gap to deployment.","safety":"Safety is mainly simulation-based validation before deployment, plus open content guardrails and SynthID watermarking across the Cosmos models; there is no single unified safety layer across the stack."}},{"id":"deepmind-genie","name":"DeepMind Genie","pattern":"Generative world model","essence":"A line of foundation world models that generate interactive, action-controllable, explorable environments from a prompt or image, used to simulate worlds and to train and evaluate embodied agents.","docUrl":"https://deepmind.google/models/genie/","runtimeNote":null,"byDimension":{"perception":"Genie takes a prompt image (Genie 2) or text (Genie 3) plus the user or agent's navigation actions; it synthesises a world frame by frame rather than sensing a real scene.","worldModel":"Genie is a foundation world model: it generates action-controllable, explorable environments and lets an agent see how the world evolves as it acts.","simulation":"The model is the simulator: it generatively simulates interactive worlds (Genie 3 runs in real time at 720p and 24fps, consistent for a few minutes) rather than running a physics engine or digital twin.","policyControl":"Genie provides the environment, not the controller; a separate embodied agent (DeepMind's SIMA) is given goals and drives the world with navigation actions.","simToReal":"Not a stated focus: Genie is used to train and evaluate agents in generated worlds (Genie 3 was shown evaluating the SIMA agent), with no claim of transferring policies onto real hardware.","safety":"Genie 3 was released only as a limited research preview to a small cohort of academics and creators while DeepMind develops safety mitigations and gathers feedback."}},{"id":"meta-vjepa","name":"Meta V-JEPA","pattern":"Latent-space video world model","essence":"A self-supervised video world model that learns by predicting in a latent representation space (not pixels); an action-conditioned variant enables zero-shot robot planning and control from goal images.","docUrl":"https://ai.meta.com/vjepa/","runtimeNote":null,"byDimension":{"perception":"A self-supervised video encoder trained on over a million hours of video turns raw video into embeddings that capture the semantics of the observed world, with no human labels.","worldModel":"V-JEPA 2 is a world model that predicts in an abstract latent representation space rather than reconstructing pixels, learning how objects move and interact.","simulation":"Not a simulator or digital twin: it predicts future latent states, not pixel-level rollouts or a simulated environment.","policyControl":"An action-conditioned variant (V-JEPA 2-AC), post-trained on about 62 hours of robot data, enables zero-shot planning and control by specifying goals as images.","simToReal":"Meta reports deploying the model zero-shot on real robot arms for reaching, grasping, and pick-and-place (65-80% on new objects) without a simulator in the loop.","safety":"No safety or guardrail mechanism is described in the official pages; the stated constraints are research limitations such as single-time-scale prediction."}},{"id":"world-labs","name":"World Labs","pattern":"Generative 3D world models","essence":"A spatial-intelligence company building generative world models that turn a prompt, image, or video into persistent, explorable 3D worlds; focused on generation and reconstruction rather than robot control. (Press often calls these 'Large World Models'; the company's own term is 'world models'.)","docUrl":"https://www.worldlabs.ai/about","runtimeNote":null,"byDimension":{"perception":"Its models take multimodal input (a single image, text, video, panoramas, or multi-view) and reconstruct 3D structure, capturing layout, depth, lighting, and spatial form.","worldModel":"The core offering is generative world models that produce persistent, spatially consistent, explorable 3D worlds (Marble, and the real-time research preview RTFM).","simulation":"Marble can reconstruct, generate, and simulate 3D worlds and export Gaussian splats, meshes, and low-fidelity collider meshes for coarse physics; outputs are 3D worlds and assets, not an industrial digital twin.","policyControl":"Documented control is over the camera and viewpoint; the company's own taxonomy treats a planning-capable, action-producing world model as still aspirational, not shipped.","simToReal":"No sim-to-real transfer, robot deployment, or real-world policy validation is claimed in the official posts.","safety":"The reviewed official pages do not document safety, guardrails, or usage-policy commitments."}}],"decisionGuide":[{"need":"A full pipeline to simulate, train, and deploy real robots","id":"nvidia-physical-ai","entry":"NVIDIA"},{"need":"Generate interactive worlds to train and evaluate embodied agents","id":"deepmind-genie","entry":"DeepMind Genie"},{"need":"Learn a predictive world model from video and act zero-shot on a robot","id":"meta-vjepa","entry":"Meta V-JEPA"},{"need":"Generate explorable 3D worlds and assets from an image or prompt","id":"world-labs","entry":"World Labs"}]},{"id":"coding-tools","title":"Agentic coding tools","thesis":"Agentic coding tools are mapped across six dimensions - surface, autonomy, context, execution, extensibility, and control - that capture where a tool runs, how much it acts on its own, how it knows your codebase, what it can do, how you extend it, and how its actions are reviewed and bounded. These tools change quickly, so durable model versions are left to the live docs.","dimensions":[{"id":"surface","name":"Surface","summary":"Where it runs","description":"The tool's form factor: a terminal CLI, an IDE extension, a standalone editor, an async cloud agent, or the browser."},{"id":"autonomy","name":"Autonomy","summary":"How much it acts on its own","description":"Where it sits on the spectrum from inline completion, to chat, to an autonomous multi-step agent that plans and executes."},{"id":"context","name":"Context","summary":"How it knows your codebase","description":"Codebase awareness and memory: indexing and retrieval, the working context, and rules or memory files."},{"id":"execution","name":"Execution","summary":"What it can do","description":"The concrete actions it can take: edit files, run terminal commands, run tests, drive a browser, and make multi-file changes."},{"id":"extensibility","name":"Extensibility","summary":"How you extend it","description":"Ways to extend it: MCP servers, subagents, plugins or skills, custom commands, rules, and hooks."},{"id":"control","name":"Control","summary":"Review, permissions, guardrails","description":"How its actions are reviewed and bounded: approvals, sandboxing, checkpoints, audit, and human-in-the-loop."}],"entries":[{"id":"claude-code","name":"Claude Code","pattern":"Agentic CLI + IDE coding tool","essence":"Anthropic's agentic coding tool, powered by Claude models, that reads a codebase, edits files, runs commands, and integrates with development tools across the terminal, IDEs, a desktop app, and the browser. The same engine, CLAUDE.md, settings, and MCP servers work across every surface.","docUrl":"https://code.claude.com/docs/en/overview","runtimeNote":null,"byDimension":{"surface":"Runs as a terminal CLI, a VS Code extension (and Cursor), a JetBrains plugin, a standalone desktop app, and in the browser; the same engine, CLAUDE.md, settings, and MCP servers work across surfaces, and it also integrates with GitHub Actions / GitLab CI, Slack, and Chrome.","autonomy":"An autonomous multi-step agent: from a single natural-language request it plans, edits across multiple files, runs commands, and verifies results rather than offering inline autocomplete. Plan mode, subagents, background / cloud agents, and scheduled routines extend longer-running autonomy.","context":"Uses agentic search to understand the whole codebase without manual context selection. Persistent context comes from CLAUDE.md files (managed, user, project, and local scopes loaded every session) and auto memory it writes per project; rules files allow path-scoped instructions.","execution":"Reads, edits, and writes files, makes coordinated multi-file changes, runs terminal commands, runs tests and builds, and uses git to commit, branch, and open PRs. Chrome integration enables debugging live web apps, and MCP adds external tool actions.","extensibility":"Connects external tools via MCP servers (streamable HTTP or local stdio), and supports subagents, skills and custom slash commands, hooks at many lifecycle events, plugins and marketplaces, and the Agent SDK for building custom agents.","control":"A tiered permission system with allow / ask / deny rules (deny-first) managed via /permissions, plus permission modes (default, accept edits, plan, and bypass). The default mode prompts for approval before edits and commands; PreToolUse hooks can block calls, OS-level sandboxing restricts Bash, and managed policy settings give enterprises centralized control."}},{"id":"openai-codex","name":"OpenAI Codex","pattern":"Coding agent, everywhere you code","essence":"OpenAI's agentic coding tool that reads, edits, and runs code, available as an open-source terminal CLI, an IDE extension, and a cloud / async agent, with GitHub and Slack integrations and a ChatGPT app surface. Its docs recommend a current OpenAI coding model rather than fixing a default.","docUrl":"https://developers.openai.com/codex/","runtimeNote":null,"byDimension":{"surface":"Multi-surface: an open-source terminal CLI built in Rust (via npm, Homebrew, or an install script), an IDE extension for VS Code (and Cursor / Windsurf), a cloud / web agent, plus GitHub and Slack integrations and a ChatGPT app surface.","autonomy":"An autonomous multi-step coding agent that can read, edit, and run code locally or in the cloud; the cloud agent runs tasks in the background and in parallel in its own environment and can be delegated work via @codex on GitHub.","context":"Reads repository context in the working directory and layers AGENTS.md instruction files (a global file, the project root, and nested directories) into a precedence-ordered prompt chain; cloud runs let you define repositories and setup steps.","execution":"Edits files, runs terminal commands, and runs and tests code across multiple files within the selected workspace; cloud environments can be configured with tools and admin-gated internet access, and recent releases add browser use via the Chrome DevTools Protocol.","extensibility":"Acts as an MCP client and can run as an MCP server (MCP support is experimental), configured through its config file; supports built-in slash commands, subagent threads, and lifecycle hooks.","control":"Human-in-the-loop is enforced by combining a sandbox mode (read-only, workspace-write, or full access) with an approval policy, exposed as Read Only, Auto, and Full Access presets and managed via the /permissions command; full access requires an explicit bypass flag, and admins control cloud internet access."}},{"id":"cursor","name":"Cursor","pattern":"AI-native editor (VS Code fork)","essence":"An AI-native code editor by Anysphere, built as a fork of VS Code, that combines Tab autocomplete, AI chat, and an autonomous multi-step Agent mode. It is multi-model, letting you pick the model behind chat and the agent.","docUrl":"https://docs.cursor.com","runtimeNote":null,"byDimension":{"surface":"A standalone AI-native desktop IDE built as a fork of VS Code, so it retains VS Code extensions, themes, and keybindings. It also offers Cursor web / desktop cloud agents, a CLI, and integrations via Slack, GitHub, and Linear.","autonomy":"Spans the full range: Tab inline autocomplete, conversational chat, and an Agent mode that autonomously plans and executes multi-step tasks with many tool calls. Cloud Agents extend this to asynchronous, parallel, remote execution.","context":"Indexes the codebase into vector embeddings for semantic retrieval, re-synced periodically and respecting .gitignore / .cursorignore. Persistent context and memory come from Rules (.cursor/rules files, user and team rules, and AGENTS.md) included in model context.","execution":"The agent edits multiple files, runs and monitors terminal commands, performs codebase and web search, reads images, and controls a browser to screenshot and test apps. Cloud Agents run full dev environments in VMs and open PRs.","extensibility":"First-class MCP support via .cursor/mcp.json (stdio, SSE, and streamable HTTP, with OAuth). Extended by the VS Code extension ecosystem, Rules, AGENTS.md, a CLI, skills, and a Cloud Agents API; multi-model with a model picker.","control":"Checkpoints (automatic snapshots you can preview and restore), terminal sandboxing with auto-review, allowlist, and run-everything modes, configurable permissions for deletes, dotfiles, browser, and external files, plus enterprise-enforced policies. Cloud Agent output is reviewed as PRs."}},{"id":"github-copilot","name":"GitHub Copilot","pattern":"Completion to autonomous coding agent","essence":"GitHub's AI coding assistant spanning IDE code completion, chat, in-editor agent mode, an asynchronous cloud coding agent, a CLI, and integrations on github.com and mobile. It is multi-model with a model picker.","docUrl":"https://docs.github.com/en/copilot","runtimeNote":null,"byDimension":{"surface":"Runs as IDE extensions (VS Code, Visual Studio, JetBrains, Vim / Neovim, Xcode, Eclipse, and Zed), as a Copilot CLI in the terminal, natively on github.com, and in GitHub Mobile. Surfaces range from a local editor plugin to a fully cloud-hosted async agent.","autonomy":"Spans the full spectrum: inline code completion, then Copilot Chat (Q&A), then an in-IDE agent mode that determines files, proposes edits and terminal commands, and iterates, up to the asynchronous Copilot coding agent that researches a repo, plans, makes changes on a branch, and opens a PR.","context":"Standing context comes from repository custom instructions in .github/copilot-instructions.md and reusable Markdown prompt files; chat can reference code and repositories, and Copilot Spaces act as a shared source of truth from docs and repos. The docs describe codebase awareness without committing to a persistent semantic index.","execution":"Agent mode edits multiple files, proposes and runs terminal commands, and iterates to fix errors in the local environment; the cloud agent runs in an ephemeral GitHub Actions environment where it can explore code, make changes, run tests and linters, then open a PR.","extensibility":"Supports MCP across IDEs, the CLI, the Copilot app, and github.com agents, plus a first-party GitHub MCP server and a curated MCP registry; extensibility also includes custom instructions and prompt files. Multi-model with a model picker.","control":"Human-in-the-loop is preserved: agent-mode edits and terminal commands are proposed for user approval, and the cloud agent's changes land on a branch and PR that a human reviews before merge. Repository rulesets and branch protections constrain the cloud agent, and Copilot code review adds AI review on PRs."}},{"id":"google-antigravity","name":"Google Antigravity","pattern":"Agent-first development platform","essence":"Google's agent-first development platform: a standalone IDE where AI agents plan, execute, and verify software tasks across the editor, terminal, and browser, with an Agent Manager surface for orchestrating them. At launch (Nov 2025) it shipped in free public preview with a lineup including Gemini 3 Pro, Claude Sonnet 4.5, and GPT-OSS.","docUrl":"https://antigravity.google","runtimeNote":null,"byDimension":{"surface":"A standalone cross-platform desktop application (an agent-first, VS Code-style IDE) for macOS, Windows, and Linux, presenting two surfaces: an Editor View and an agent-orchestration Manager surface; agents also drive the terminal and a Chrome browser. Launched in public preview in November 2025.","autonomy":"Spans the full range: inline tab completion and inline commands in the Editor View, up to autonomous multi-step agents that plan, execute, and verify complex tasks end-to-end and run asynchronously in the Manager surface.","context":"Codebase-aware agents operating within workspaces; it treats learning as a core primitive, letting agents save useful context and code snippets to a knowledge base to improve future tasks.","execution":"Agents act across the editor (multi-file code edits, tab completions, and inline commands), the terminal (for example launching the app), and the browser (testing and verifying via Chrome, producing browser recordings).","extensibility":"Supports MCP with local and remote servers (stdio, SSE, and streamable HTTP) via an MCP Store with one-click Google Cloud integrations and a user-editable config file, plus per-project control over which MCP tools are exposed to the agent.","control":"Human-in-the-loop via Artifacts (task lists, plans, screenshots, and browser recordings) reviewed at a glance with Google-Docs-style comments the agent incorporates without halting; the permissions docs describe an Ask mode with allow / deny rules, and browser actions require explicit Chrome debugging permission."}}],"decisionGuide":[{"need":"A terminal-first agent with deep memory and extensibility","id":"claude-code","entry":"Claude Code"},{"need":"One agent across CLI, IDE, and the cloud","id":"openai-codex","entry":"OpenAI Codex"},{"need":"An AI-native editor with fast tab completion and agent mode","id":"cursor","entry":"Cursor"},{"need":"Agentic coding inside GitHub, from completion to PRs","id":"github-copilot","entry":"GitHub Copilot"},{"need":"A standalone agent-first IDE built around Gemini","id":"google-antigravity","entry":"Google Antigravity"}]}]}