For the complete BrainAPI documentation index, see llms.txt. A markdown version of any docs page is available by appending .md to its URL. Docs MCP: /docs/mcp.
Choose a path
Match your application, operating, or extension goal to the shortest BrainAPI workflow
You do not need to learn every BrainAPI surface before building something useful. Choose the outcome that resembles your problem, complete the smallest workflow, and add graph, ranking, personalization, or agents only when evidence justifies the extra cost.
Build an application
Ground an assistant in private knowledge
Ingest documentation, tickets, or conversation summaries, then call /retrieve/context for a compact context pack. Choose this path when the next component is an LLM and it needs both relevant passages and connected facts.
Start with Quickstart, then read Context.
Add ranked search
Use /retrieve/search when your interface displays ordered results, snippets, facets, or score explanations. Core hybrid passages cover most documentation and support corpora. Graph channels, plugins, and personalization are optional levels rather than prerequisites.
Start with Search levels and copy a complete Search recipe.
Recommend the next item
Use /retrieve/recommend when the request starts from a user, course, media asset, document, or product rather than a text query. BrainAPI can use direct graph relationships without training a model; RecSys GNN is an additive plugin when you have enough interaction data to justify training.
Start with Recommendations.
Operate an instance
Use the TUI for the shortest local setup or Installation for explicit development and production flows. Keep Configuration nearby when choosing model and storage backends. When a service is running but requests fail, use the symptom-led troubleshooting guide.
Extend the platform
Use MCP when an external agent needs existing BrainAPI capabilities. Build a plugin when you need a new route, lifecycle hook, ingestion behavior, MCP tool, first-stage search retriever, or second-stage reranker.
Choose the lightest retrieval surface
| Your caller needs | Surface | Avoid it when |
|---|---|---|
| Prompt-ready evidence | Context | You need a stable ranked result list or facets. |
| Ranked hits | Search | You only need context for one LLM call. |
| Related items or next actions | Recommendations | The request is fundamentally a text query. |
| Exact graph state or traversal | Graph APIs | A passage-level answer is enough. |
| Multi-step investigation | MCP and agent orchestration | One bounded retrieval request answers the question. |
Higher levels are not automatically better. Measure the failure you are fixing before adding graph reads, a plugin index, reranking, personalization, or agent orchestration.
Last updated on