Search Backends

Choosing the right similarity search engine for your organization

Search Backends

Simili Bot v0.2.0 offers a flexible search layer. You can choose from three distinct backends depending on your needs for speed, accuracy, and infrastructure overhead.

Backend Comparison

Feature	GitHub Native	Qdrant	BM25
Setup Type	Zero-Config	Advanced	Built-in
Search Type	Hybrid (Semantic + Keyword)	Dense Vector (Semantic)	Sparse Vector (Keyword)
External VDB	Not Required	Required	Not Required
Embedding API	Not Required	Required	Not Required
Accuracy	High	Very High	Medium
Scale	Repo-level	Organization-wide	Repo-level

---

1. GitHub Native (Recommended for Start)

The github_native backend uses GitHub’s internal search infrastructure. It is a Zero-Config option that requires no external databases or embedding API keys for the search phase.

Best for: Most teams starting with Simili Bot.

search:
  backend: "github_native"
  bm25_fallback: true

Note: While search is zero-config, the LLM triage and verification steps still require a GEMINI_API_KEY or OPENAI_API_KEY.

2. Qdrant (Recommended for Scale)

The qdrant backend uses a dedicated vector database. It provides the highest accuracy for deep semantic relationships and is designed for organization-wide discovery.

Best for: Large organizations with thousands of issues across many repositories.

search:
  backend: "qdrant"

Requirements:

• A Qdrant instance (Cloud or Self-hosted).
• An embedding provider (Gemini or OpenAI) configured in the embedding: block.

3. BM25 (Keyword-Based)

BM25 is a classic ranking function used by search engines to estimate the relevance of documents to a given search query.

Best for: Finding exact error codes, specific technical terms, or as a local fallback.

search:
  backend: "bm25"

---

Hybrid Search & Fallbacks

Simili Bot allows you to combine backends for maximum reliability.

BM25 Fallback

Enable bm25_fallback to ensure that if the primary backend (Qdrant or GitHub Native) fails or returns zero results, the bot performs a local keyword search.

search:
  backend: "github_native"
  bm25_fallback: true

Cold Start with learn

If you are using the Qdrant backend for routing, use the simili learn command to index your repository documentation. This helps the bot route issues correctly even when no historical issues are present.

Learn Command

Discover how to index repo docs for semantic routing.