Skip to main content
Hiroshi features a file-backed Memory Wiki system that chunks local markdown manuals, generates vector embeddings, and performs fast mathematical cosine similarity lookups.

​
⚙️ 1. Configurations Setup

To enable Memory Wiki search, update settings inside config.toml: 
[wiki]
enabled = true
wiki_dir = "~/.hiroshi/workspace/wiki"
similarity_threshold = 0.75

​
🛠️ 2. Execution Primitives

​
<wiki_store>

Index or re-index the markdown knowledge base: 
<wiki_store>/path/to/custom/wiki</wiki_store>
  • Slices markdown blocks at paragraph boundaries (\n\n).
  • Caches generated vector embeddings in-memory.

​
<wiki_search>

Search semantic vector databases using cosine dot-product similarity: 
<wiki_search>how to deploy on digitalocean</wiki_search>
  • Generates query embedding vectors.
  • Performs dot-product calculation matching chunks with scores exceeding similarity_threshold.