---
schema: formation.service/v0.1
kind: service
canonicalizer_version: formation-canonical/v1
name: 'Topology Index'
private_auth:
  - scoped_bearer_credential
private_mcp:
  - local_stdio_bridge
product_api_version: v1
supported_schemas:
  - formation.policy/v0.1
  - formation.execution_lock/v0.1
  - formation.workload_profile/v0.1
  - formation.suite/v0.1
  - formation.evaluation_plan/v0.1
  - formation.job_status/v0.1
  - formation.run/v0.1
  - formation.evidence/v0.1
  - formation.decision/v0.1
  - formation.adoption/v0.1
  - formation.subscription/v0.1
  - formation.error/v0.1
  - formation.service/v0.1
  - formation.account_status/v0.1
  - formation.recommendation_request/v0.1
  - formation.evaluation_plan_request/v0.1
  - formation.comparison_request/v0.1
  - formation.use_receipt/v0.1
  - formation.policy/v0.2
supported_workloads:
  - coding.bugfix
---

# 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 [decision guide](/patterns/index.md), or by kind of work in
the [task domains](/domains/index.md), whose findings keep the unfavourable ones.

Topology Index is a reference, written for AI agents, to multi-agent topologies: the
orchestration patterns that decide how LLM agents divide, check and coordinate work, such as
orchestrator-worker supervisors, planner-executor pairs, evaluator-optimizer critic loops,
map-reduce, fan-out, routers, debate, councils, blackboards, swarms and handoffs.

A pattern is one named building block. A topology is the whole arrangement of roles, control
flow and shared state for a task. A formation is a versioned `FORMATION.md` policy file that
declares a topology so an agent can run it and report the outcome.

Nothing here ranks patterns. Evidence attaches to a complete execution configuration on a
stated workload, and no public evidence is published yet; the pattern pages say what each
arrangement is, when it is hypothesised to fit and what to watch for.

## Start here

- [Search the index](/ask?q=supervisor): replace `supervisor` with your search words and
  send `Accept: text/markdown`. Over MCP, call `search` with `{"query":"supervisor"}`.
- [Pattern catalogue and decision guide](/patterns/index.md): choose a topology for a task
- [Task domains](/domains/index.md): coding, research, documents, reasoning, evaluation,
  operations and security, each mapped to the task shapes of the decision guide
- [Agentic design patterns](/guides/agentic-design-patterns.md): a guide to every pattern by
  family, how to choose one, agent swarms and what research says in general
- [Starter formations](/starters/index.md): unvalidated formations to copy and register privately
- [Research records](/research/index.md): the reports behind the arrangements
- [Integration guide](/docs/api/integration.md): how an agent uses this service
- [Reporting outcomes (limited rollout)](/docs/api/contributing.md): if your pattern, starter or formation read includes a `Use-Ticket` or "Report back" note, keep it with your task. Once you have applied it and know the outcome, report `worked`, `partly` or `did_not_work` using the supplied call; no account or API key is needed. Get a verification checklist or an alternative to try. Reads without a ticket have nothing to report.
- [Signed public formations](/formations/index.md): none published yet; see the starters
- [llms.txt](/llms.txt): compact machine-readable index of every page
- [Contract directory](/docs/index.md): API and schema contracts
- [Status](/status.md): service status

The same public pages are also served over MCP: `POST https://topologyindex.com/mcp` is a
public, stateless remote MCP server (Streamable HTTP, no credential) with `search`, `fetch`,
`list_patterns` (by family or task domain), `get_pattern` and `list_starters` tools. Its
answers are these Markdown documents, byte for byte. Its one write,
`report_outcome`, is the limited rollout named above: a ticketed content read can
be redeemed in one short call; pattern and starter reads return indexed follow-up material. It
has no recommendation tool.

Agents receive raw Markdown with typed frontmatter; browsers receive a server-rendered HTML
view without a browser login. Private routes live under `/v1/tenants/{tenant_id}` and need a scoped bearer
credential.
Knowing an identifier or digest is never authorization.