Vistara Local file search engine

Major update

No more waiting for hugginface model to use Vistara

its now bundled with Vistara 🥳

Vistara Local file Search engine

QoL improvements I had made

• Open file directly from search -You can open the file directly from vistara. no need to Go to the Freaking path
• Changed file Detection -If a file is changed, It is treated as Changed . Not New file.
• Sensitive files detection -To avoid exposing contents like Passwords,Api keys,Tokens and even Address of ur loved ones

VISTARA - Day 5

ready for ship

added UI !!!!

VISTARA — Day 4 Devlog

Date: August 18, 2026
Focus: Semantic Search + Hybrid Search

Today I expanded VISTARA from keyword-based search into semantic search, allowing it to understand the meaning behind queries.

Completed

• Sentence Transformers integration
• all-MiniLM-L6-v2 embedding model
• 384-dimensional embeddings
• Embedding generation and storage
• Cosine similarity search
• Semantic result ranking
• Compact semantic snippets
• Hybrid keyword + semantic search
• Sensitive-file filtering in semantic search
• Combined FTS5 and semantic results

Result

VISTARA can now search using both exact keywords and semantic meaning, making the search much more useful for natural-language queries.

The search engine backend is now essentially complete.

Next

Move towards the remaining polish and then Implement User Interface.

Day 4: Completed ✅

Semantic Search coming soon

Vistara — Day 3 Devlog

Date: August 17, 2026
Focus: Keyword Search + Chunking

Today I expanded Local Google from basic keyword search into chunk-based search.

Completed

• SQLite FTS5 keyword search
• BM25 relevance ranking
• Sensitive-file filtering
• Search result formatting
• 500-word text chunking
• chunks database table
• chunks_fts FTS5 index
• Backfilled chunks for existing documents
• Chunk-based keyword search
• Search-result snippets from matching chunks
• Best matching chunk per document to reduce duplicates

Result

Local Google can now search through document chunks instead of treating every document as one giant piece of text.

The original Day 4 chunking work was moved into Day 3 because most of it was completed today.

Next

Finish the remaining ranking polish, then move toward semantic search with embeddings.

Day 3: Completed ✅

VISTARA — Day 2: File Ingestion 🔎

VISTARA is getting smarter today.

It started as a file scanner that could tell me where my files are. Now it’s learning how to actually read what’s inside them. 🗿

VISTARA can currently process:

• .txt
• .md
• .pdf
• .docx

It can scan across the PC while skipping unnecessary system and development folders, keeping the focus on useful documents.

It also uses SHA-256 hashing to detect whether a document is:

• 🆕 New
• 🔄 Changed
• 😴 Unchanged

The first full scan found 7,462 supported files. 💀

Now VISTARA is going through them one by one, extracting their contents and storing the information in SQLite.

Basically:

"Where tf did I save that file?"

becomes:

"VISTARA, find it." 🗿

And yes… VISTARA is currently eating thousands of files, so my laptop is getting its own workout. 💀🔥

Next up: SQLite FTS5 keyword search.

VISTARA — a local file search engine built to make finding things on your PC less annoying.

First Look Scanning on progress

Project Skeleton Day 1