example

README

Examples

Runnable, self-contained programs showing how to use recall. Every example is deterministic and offline: they use embedder.NewMockEmbedder and llm.NewMockBackend, so they need no API keys, network access, or configuration. Run any of them from the repo root:

go run ./example          # quick tour: upload, search, hybrid, graph, graph-RAG
go run ./example/e2e      # full lifecycle: ingest -> search -> RAG -> eval -> reasoning
go run ./example/production  # API server deployment driven by the typed client

What each example covers

Program	Demonstrates
main.go	The 30-second tour: store.MemoryStore upload + vector search, hybrid search, graph.KnowledgeGraph (entities, relations, path finding), and graph-based extraction via store.NewMemoryGraphStore.
e2e/main.go	The end-to-end RAG tutorial: directory-loader ingestion through ingest.Pipeline (dedup + progress) into a SQLite store, vector vs. hybrid search, the RAG pipeline with citations answered by a mock LLM, retrieval evaluation (Precision/Recall/MRR/NDCG@K) with eval.BenchmarkSuite, and knowledge-graph extraction + multi-hop reasoning.
production/main.go	Service deployment: app.BuildAPIServer (the same assembly the recall-server binary uses — SQLite store, RAG pipeline, graph, reasoner) served over HTTP on an ephemeral port, driven entirely through the typed client package (Health, Upload, Search, RAG, Diagnostics), with graceful shutdown.

From examples to production

• Real embeddings: replace embedder.NewMockEmbedder(384) with embedder.NewOpenAIEmbedder, NewCohereEmbedder, NewOllamaEmbedder, or the local ONNX embedder. API keys come from environment variables — never hardcode them.
• Real LLMs: replace llm.NewMockBackend with llm.NewOpenAIClient or llm.NewOllamaClient; wrap with llm.NewRetryBackend / NewRateLimitBackend / NewCircuitBreakerBackend for resilience.
• Real servers: run the standalone recall-server -config recall.yaml (see SECURITY.md for API-key auth) and point the client (or the recall CLI) at it.

Benchmark comparison

For index/algorithm comparisons see docs/BENCHMARKS.md and scripts/benchcompare.sh, which runs the benchmark suite and diffs against a stored baseline.

Documentation

Overview

Package main demonstrates common usage patterns for the recall library.