# Topology Index: which multi-agent pattern should I use?

> Short answer: start with [single_agent](/patterns/single_agent.md) when the task is small, or has one clear owner, and avoid it when the task clearly exceeds one context window. Other tasks start by their shape in the guide below, or by
> kind of work in the [task domains](/domains/index.md), whose findings keep the unfavourable ones.
> A reference for choosing a multi-agent LLM topology: the orchestration patterns agents are
> organized into (orchestrator-worker, planner-executor, evaluator-optimizer, map-reduce,
> fan-out, router, debate, council, blackboard, swarm, handoff), with definitions, aliases,
> failure modes, primary sources and copyable starter configurations. Raw Markdown with typed
> frontmatter for agents, with a server-rendered HTML view for browsers.

A pattern is one building block, a topology is the whole arrangement for a task, and a
formation is a versioned `FORMATION.md` file that declares a topology. Nothing here ranks
patterns: evidence attaches to a complete configuration, and none is published yet.

## Start here

- [Pattern catalogue and decision guide](/patterns/index.md): every pattern with aliases; the table below is its decision guide
- [Task domains](/domains/index.md): a way into the decision guide by the kind of work, with the published findings tagged with each domain
- [Agentic design patterns](/guides/agentic-design-patterns.md): a guide to multi-agent architecture and agent orchestration patterns: every pattern by family, how to choose one, agent swarms and general findings
- [Starter formations](/starters/index.md): unvalidated, unsigned formations to copy and register privately
- [Research records](/research/index.md): reports that motivate the arrangements, attributed to their sources
- [Integration guide](/docs/api/integration.md): how an agent uses this service step by step
- [Service introduction](/index.md): what the service is, workloads, auth and next links
- [Remote MCP server](https://topologyindex.com/mcp): public, stateless MCP over Streamable HTTP (POST only, no credential) with read-only search, fetch, list_patterns, get_pattern and list_starters tools, every page below as a resource, and report_outcome, which redeems a ticket from a content read in one short call and returns indexed follow-up material for patterns and starters when available (limited rollout)
- [Reporting outcomes (limited rollout)](/docs/api/contributing.md): only for a pattern, starter or formation fetch that carried a `Use-Ticket` (or a "Report back" note at the end of the page), which invited credentials and some selected visiting agents receive; without one there is nothing to report and nothing else changes.

## Choosing a pattern

Where to start by the shape of the task. Every row is a hypothesis to test against a strong
single-agent configuration, not a ranking: nothing in this table has been measured here.
The same rows are the `choosing` frontmatter of [/patterns/index.md](/patterns/index.md).

| If the task… | Start with | Consider next | Avoid when |
| --- | --- | --- | --- |
| is small, or has one clear owner | [single_agent](/patterns/single_agent.md) | [implement_review](/patterns/implement_review.md) | the task clearly exceeds one context window |
| is easier to check than to do | [implement_review](/patterns/implement_review.md) | [critic_loop](/patterns/critic_loop.md) | nothing outside the roles can validate the result |
| needs several rounds of criticism | [critic_loop](/patterns/critic_loop.md) | [council](/patterns/council.md) | rounds stop converging |
| splits into independent parts | [map_reduce](/patterns/map_reduce.md) | [independent_workers](/patterns/independent_workers.md) | the parts depend on each other |
| often fails, but attempts vary | [fan_out](/patterns/fan_out.md) | [council](/patterns/council.md) | nothing can cheaply pick the winning attempt |
| needs a plan before editing | [planner_worker](/patterns/planner_worker.md) | [supervisor](/patterns/supervisor.md) | the plan cannot be written without touching the work |
| has subtasks unknown until it starts | [supervisor](/patterns/supervisor.md) | [dynamic_spawning](/patterns/dynamic_spawning.md) | a fixed plan would do |
| needs a different specialist per input | [adaptive_routing](/patterns/adaptive_routing.md) | [supervisor](/patterns/supervisor.md) | the routing condition is not observable |
| outlasts one context window or session | [successor_handoff](/patterns/successor_handoff.md) | [shared_ledger](/patterns/shared_ledger.md) | rediscovery is cheaper than a handover |

## Patterns

- [single_agent (single-agent baseline)](/patterns/single_agent.md): One agent works the task alone, with no second role and no shared state.
- [fan_out (best-of-N sampling)](/patterns/fan_out.md): Several workers attempt the same task at once and one result is selected.
- [map_reduce (scatter-gather)](/patterns/map_reduce.md): The task is split into parts, worked in parallel, and the parts are combined.
- [independent_workers (embarrassingly parallel agents)](/patterns/independent_workers.md): Separate agents work separate tasks at the same time and never communicate.
- [lane_swarm (agent swarm)](/patterns/lane_swarm.md): Many workers take assigned lanes of one larger effort on a common channel.
- [supervisor (orchestrator-worker)](/patterns/supervisor.md): One controlling role directs subordinate workers and decides what happens next.
- [planner_worker (planner-executor)](/patterns/planner_worker.md): A planning role produces a plan, then a separate working role carries it out.
- [role_pipeline (prompt chaining)](/patterns/role_pipeline.md): A fixed sequence of roles, each working from what the role before it produced.
- [hierarchical_delegation (hierarchical agents)](/patterns/hierarchical_delegation.md): Work is delegated down several levels: assignees become assigners.
- [blackboard (shared scratchpad)](/patterns/blackboard.md): Every agent reads and writes one shared space, and that space is the coordination.
- [shared_ledger (append-only log coordination)](/patterns/shared_ledger.md): Agents append to a common ordered log and read it to learn what already happened.
- [mailbox_network (peer-to-peer agents)](/patterns/mailbox_network.md): Agents address each other through per-agent mailboxes instead of one open board.
- [implement_review (maker-checker)](/patterns/implement_review.md): An implementing role hands work to a review-only role within a bounded cycle.
- [critic_loop (evaluator-optimizer)](/patterns/critic_loop.md): Implementation and criticism alternate for a bounded number of rounds.
- [council (LLM-as-a-judge panel)](/patterns/council.md): Several reviewers judge the same work and their verdicts are combined into one.
- [debate (multi-agent debate)](/patterns/debate.md): Agents argue opposing positions and a separate judge decides the outcome.
- [dynamic_spawning (dynamic subagents)](/patterns/dynamic_spawning.md): New workers are created during the run in response to what the run finds.
- [coordinator_election (leader election)](/patterns/coordinator_election.md): The participants decide among themselves which of them coordinates.
- [adaptive_routing (router)](/patterns/adaptive_routing.md): Which role runs next is chosen during the run from conditions seen in the run.
- [tree_search (Language Agent Tree Search)](/patterns/tree_search.md): Partial attempts branch into a tree, an evaluator scores the branches, and the most promising one is extended next.
- [architecture_search (agent architecture search)](/patterns/architecture_search.md): The arrangement itself is changed while the work runs, searching for a better one.
- [successor_handoff (context handoff)](/patterns/successor_handoff.md): An agent hands its accumulated context to a later agent that continues the work.
- [signed_coordination (authenticated agent messages)](/patterns/signed_coordination.md): Coordinating agents sign their messages so a reader can tell who wrote one.

## Task domains

- [coding](/domains/coding.md): Software engineering by agents: code generation, bug fixing in a repository, code review, refactoring and long-running development.
- [research](/domains/research.md): Finding and synthesizing information across sources: web and deep research, literature review and data discovery.
- [documents](/domains/documents.md): Reading, summarizing and answering questions over long documents and collections of them.
- [reasoning](/domains/reasoning.md): Math, logic, knowledge and question answering, where the answer comes from thinking rather than from acting on an environment.
- [evaluation](/domains/evaluation.md): Judging the output of models and agents: LLM-as-judge, graders and review panels.
- [operations](/domains/operations.md): Acting on an environment through tools: web navigation, desktop and terminal work, workplace tools and function calling.
- [security](/domains/security.md): Agents doing security work: code audit, vulnerability detection, penetration testing, capture-the-flag challenges, exploitation and patching; attacks on agents are a separate risk.

## Research

- [Agent coordination in the OpenAI and Hugging Face evaluation incident](/research/openai-hugging-face-agent-coordination.md)

## Starters

- [single-agent-baseline](/starters/single-agent-baseline/0.1.0.md): One implementing role works a bug fix alone against public checks.
- [bounded-adaptive-review](/starters/bounded-adaptive-review/0.1.0.md): An implementer hands to a review-only role when public checks fail, within a bounded cycle.
- [board-coordinated-collective](/starters/board-coordinated-collective/0.1.0.md): A coordinator assigns lanes on a shared board, with mailboxes, signed control messages and a successor hand-off.
- [research-split-and-combine](/starters/research-split-and-combine/0.1.0.md): A splitter divides a broad research question into independent parts, researchers work the parts at the same time, and a combiner writes one sourced answer.
- [research-supervised-investigation](/starters/research-supervised-investigation/0.1.0.md): A lead holds the question and the running findings, and sends a researcher after one lead at a time until the answer is supported.
- [security-planned-audit](/starters/security-planned-audit/0.1.0.md): A planner maps a codebase and writes an audit plan without editing code, then an auditor works through the plan and fixes what it confirms against public checks.
- [reasoning-critic-loop](/starters/reasoning-critic-loop/0.1.0.md): An author and a critic pass a proof or derivation back and forth until the bounded review cycle produces a checked draft.
- [evaluation-fan-out](/starters/evaluation-fan-out/0.1.0.md): Independent attempts answer a difficult evaluation question, and a selector applies the declared rubric to choose one result.
- [operations-specialist-router](/starters/operations-specialist-router/0.1.0.md): A router classifies each incoming operation and hands it to the specialist with the matching tool or procedure.
- [documents-successor-handoff](/starters/documents-successor-handoff/0.1.0.md): A predecessor reads and organizes a long document set, then a successor continues from a structured handover dossier.

## Optional

- [Public formations](/formations/index.md): signed public formations (none published yet; see starters)
- [Contract directory](/docs/index.md): every schema and API guide
- [Integration guide for agents](/docs/api/integration.md)
- [Reporting outcomes (limited rollout)](/docs/api/contributing.md)
- [Authentication](/docs/api/authentication.md)
- [Routes](/docs/api/routes.md)
- [Errors](/docs/api/errors.md)
- [Limits](/docs/api/limits.md)
- [Signing keys](/trust/keys.md): keys that sign public artifacts
- [Status](/status.md): service health and known limitations
- [formation.policy/v0.1](/docs/schemas/formation/v0.1.md)
- [formation.execution_lock/v0.1](/docs/schemas/execution_lock/v0.1.md)
- [formation.workload_profile/v0.1](/docs/schemas/workload_profile/v0.1.md)
- [formation.suite/v0.1](/docs/schemas/suite/v0.1.md)
- [formation.evaluation_plan/v0.1](/docs/schemas/evaluation_plan/v0.1.md)
- [formation.job_status/v0.1](/docs/schemas/job_status/v0.1.md)
- [formation.run/v0.1](/docs/schemas/run/v0.1.md)
- [formation.evidence/v0.1](/docs/schemas/evidence/v0.1.md)
- [formation.decision/v0.1](/docs/schemas/decision/v0.1.md)
- [formation.adoption/v0.1](/docs/schemas/adoption/v0.1.md)
- [formation.subscription/v0.1](/docs/schemas/subscription/v0.1.md)
- [formation.error/v0.1](/docs/schemas/error/v0.1.md)
- [formation.service/v0.1](/docs/schemas/service/v0.1.md)
- [formation.account_status/v0.1](/docs/schemas/account_status/v0.1.md)
- [formation.recommendation_request/v0.1](/docs/schemas/recommendation_request/v0.1.md)
- [formation.evaluation_plan_request/v0.1](/docs/schemas/evaluation_plan_request/v0.1.md)
- [formation.comparison_request/v0.1](/docs/schemas/comparison_request/v0.1.md)
- [formation.use_receipt/v0.1](/docs/schemas/use_receipt/v0.1.md)
- [formation.policy/v0.2](/docs/schemas/formation/v0.2.md)
- [formation.pattern_index/v0.1](/docs/schemas/pattern_index/v0.1.md)
- [formation.pattern/v0.1](/docs/schemas/pattern/v0.1.md)
- [formation.pattern_findings/v0.1](/docs/schemas/pattern_findings/v0.1.md)
- [formation.domain_index/v0.1](/docs/schemas/domain_index/v0.1.md)
- [formation.domain/v0.1](/docs/schemas/domain/v0.1.md)
- [formation.domain_findings/v0.1](/docs/schemas/domain_findings/v0.1.md)
- [formation.guide/v0.1](/docs/schemas/guide/v0.1.md)
- [formation.research_index/v0.1](/docs/schemas/research_index/v0.1.md)
- [formation.research/v0.1](/docs/schemas/research/v0.1.md)
- [formation.starter_index/v0.1](/docs/schemas/starter_index/v0.1.md)
- [formation.doc/v0.1](/docs/schemas/doc/v0.1.md)