Explore
One faceted index over all patterns, ADPs, anti-patterns, frameworks, and learning paths.
- ReActThink, act, look, think again.
- Plan-and-ExecutePlan once. Execute the list.
- ReWOOPlan. Batch. Fold.
- ReflexionAct. Evaluate. Reflect. Retry.
- Tree of ThoughtsBranch the reasoning. Keep the best leaf.
- Self-ConsistencyVote across many minds. Settle on the majority.
- CodeActThink in code. Execute. Observe. Repeat.
- Sequential PipelineFixed steps. No back-edges.
- RoutingClassify once. Dispatch to the right specialist.
- ParallelizationSplit the work. Run in parallel. Merge the results.
- LoopRepeat until good enough, or budget runs out.
- Evaluator-OptimizerGenerate. Judge. Improve. Repeat.
- Iterative RefinementOne artifact. Many passes. Each better than the last.
- Orchestrator-WorkersDecompose. Dispatch. Fold.
- Map-ReduceMap over chunks. Reduce to one answer.
- Resource-Aware OptimizationRoute by difficulty. Pay for what the task needs.
- SupervisorOne decides. Specialists do.
- Hierarchical SupervisorSupervisors of supervisors. Scale by layers.
- HandoffOne agent yields. Another takes over.
- SwarmLocal rules. Global emergence. No center.
- Group ChatA shared thread. Everyone speaks.
- Multi-Agent DebateArgue the answer. An arbiter decides.
- MagenticPlan. Delegate. Synthesize. Over long horizons.
- BlackboardA shared scratchpad. Specialists react.
- Contract NetBroadcast. Bid. Award. Execute.
- Agents-as-ToolsCall a specialist like any other tool.
- Graph-based OrchestrationAn executable graph of agents, tools, and state.
- Exploration & DiscoveryMap the space. Cluster, select, deep-dive, synthesize.
- Memory ArchitectureShort-term, long-term, retrieved.
- Tool RegistryA catalog, not a kitchen sink.
- MCPTools speak a shared protocol.
- A2A ProtocolAgents address other agents.
- CheckpointingSave state. Resume cleanly.
- Workflow DAGA graph that can be drawn.
- HITL GatePause before anything irreversible.
- Sandbox ExecutionRun code where it can't escape.
- Audit TrailEvery step, signed and timestamped.
- LLM-as-JudgeAn LLM grades another LLM.
- Distributed TracingSpans across the whole agent loop.
- Conversational MemoryThe running message history the agent reads each turn.
- Episodic MemoryPast tasks, persisted, so the agent can recognise its own déjà vu.
- Semantic / Vector / Graph MemoryWhat the agent knows, indexed for retrieval.
- Working Memory / ScratchpadPer-task notes the agent keeps while it reasons.
- Virtual Context ManagementThe agent as its own memory manager: page facts in and out like an OS.
- RecorderSnapshot the state so the run can be resumed, not restarted.
- Skill BuildTurn winning trajectories into named, reusable procedures.
- Function CallingStructured JSON in, typed function call out.
- Adapter PatternWrap the messy API in a clean tool the model can call.
- Capability RoutingPick the tool by what it can do, not what it's called.
- Permission-scoped ToolsEach tool carries the smallest permission set that lets it work.
- Agentic RAGRetrieve only when it helps — and decide that yourself.
- Actor ModelEach agent is its own mailbox, its own state, its own supervisor.
- Event-driven ChoreographyNo conductor. Each component listens for the events it cares about.
- Saga / CompensationEvery forward step ships with its own undo.
- Pub/Sub Agent MeshTopics, not addresses. Agents subscribe to what they care about.
- Output Validation / Schema EnforcementReject malformed model output before it pollutes the next step.
- Least Privilege AgentEach agent sees the smallest world that lets it do its job.
- ControllerPolices the agent's actions against policy as the run unfolds.
- IntegratorSanitise the percepts before reasoning touches them.
- Multimodal GuardrailsPer-modality filters on every input and output the agent touches.
- Statistical GuardrailsReject on the numbers: cosine-distance drift and token-entropy confidence.
- Token / Cost TrackingTokens and dollars, per run, per agent, per node.
- Integration Tests for AgentsReplayable end-to-end scenarios — fixtures, traces, evals.
- IntegratorValidates incoming observations before they enter the world model, ensuring downstream reasoning operates on clean signals rather than raw, potentially malformed inputs.
- RetrieverContext-sensitive interface to long-term memory; selects what is salient to the current step rather than dumping the entire store into the prompt.
- RecorderSaves and restores Reasoning & World Model (RWM) states for durability, replay, and forward recovery across steps, sessions, and processes.
- SelectorDynamic prioritization of goals — the tactical step-selector that picks the next action from competing demands.
- PlannerStrategic decomposition of complex goals into ordered sub-goals that can be executed sequentially or in parallel.
- DeliberatorSelection of the optimal action per planning step — the local choice given the local context, not a global default.
- ExecutorReliable execution and systematic feedback collection, with structured retry and outcome capture.
- Tool UseProxy / adapter interface for safe external function calls, enforcing argument validation and capability scoping.
- CoordinatorManagement of structured multi-agent communication, ensuring message flow is inspectable and bounded.
- ReflectorCausal failure analysis for strategy adjustment — the gap between "it failed" and "it failed because...".
- Skill BuildExtraction of reusable procedures from past experience, converting one-off solutions into persistent assets.
- ControllerContinuous monitoring of ethical and operational guardrails, catching drift before it cascades.
- Over-AgentificationHigh latency, expensive token bills, debugging that requires reconstructing emergent behaviour from logs.
- Hidden State in PromptsBehaviour drifts over long conversations. Bugs reproduce only intermittently.
- SQLite Under ConcurrencyWrite locks serialize everything. Timeouts under load. Session bleeding.
- Tool ExplosionTool selection accuracy collapses. The model hallucinates calls and parameter names.
- Hallucinated RoutingReflection loops never converge. Recursion limits trip at runtime instead of design time.
- God OrchestratorSingle point of coordination, single point of failure, single point of privacy exposure.
- Cascading Security FailuresA poisoned doc in a shared index contaminates every downstream consumer. Prompt injection propagates via handoffs.
- Unbounded LoopAn L1 agent stuck in unproductive infinite cycles due to flawed reasoning, with no programmed recursion limit or step budget.
- Self-Graded HallucinationA model that critiques its own answer can agree with its own mistake — a reflection loop converges, confidently, on something wrong.
- AWS Strands SDKModel-driven minimal SDK; the LLM weaves the graph from a 'bundle of threads' of tools.
- CrewAIRole-based 'Crews' of agents; flat abstraction over team-shaped workflows.
- AutoGen / AG2Event-driven conversation between agents; asynchronous messages and group chats.
- Google ADKModular, hierarchical workflow agents for the Vertex AI ecosystem.
- OpenAI Agents SDKLightweight, Python-first SDK for handoff-based agent flows.
- SuperAgentSingle-agent assistant platform for customer-facing deployments.
- MetaGPTSOP-based multi-agent metaprogramming that simulates an IT company.
- Pydantic AIType-safe Python agent framework; structured outputs and validation/retry loops on Pydantic v2.
- LlamaIndexRetrieval-centric framework; connectors and indexes turn proprietary documents into agent context.
- Semantic KernelEnterprise SDK for .NET/Java/Python; skills and planners under Microsoft/Azure governance.
- LangChain4jBrings LLM and agentic capabilities to the JVM (Java/Kotlin) with LangChain-style abstractions.
- LangGraphGraph-based, state-driven runtime; explicit nodes and edges, checkpointer-backed.
- TemporalDurable workflow engine for minutes-to-hours processes; deterministic replay and automatic retries.
- InngestEvent-driven durable execution; steps, retries and flow control over a queue you don't manage.
- RestateDurable execution for distributed services; durable promises, virtual objects and exactly-once handlers.
- Deep Agents SDKOpinionated, batteries-included harness on LangGraph: planning, subagents and a virtual file system out of the box.
- Claude Agent SDKAnthropic's agentic harness — the gather-context / act / verify loop behind Claude Code, with tools, subagents and MCP.
- The newcomerStart here if multi-agent systems are new to you.
- The pattern pickerYou have a real task and need a defensible pattern choice.
- The framework chooserYou are evaluating frameworks and want to compare with eyes open.
- The operatorYou will run this in production and want a checklist with named failure modes.
- The threat modelerYou are securing a multi-agent system and want the attack surface, the defenses, and a worked example — in that order.