AstraClaw AI is KeepCloud's built-in intelligence layer — a multi-agent search system that understands natural language queries (including Hindi and Hinglish), reads inside files using OCR, recognizes document types by their content, and delivers high-precision results across all your connected clouds.
Finding a specific file when you can't remember its name is one of the most common and frustrating experiences in digital organization. You know you have a scan of your electricity bill from February, but was it uploaded to Drive or Dropbox? Was it named "bill_feb_2026" or something like "IMG_20260215_115430"? Is it a PDF, a JPG, or a Word document?
AstraClaw AI was built to solve this. Not by guessing — but by actually reading your files and understanding what they contain.
What makes AstraClaw AI different
Most cloud search tools search filenames only. Type "electricity bill" and you get files that literally have those words in their name. If your bill is named "IMG_20260215_scan.jpg", a name-only search returns nothing.
Astra goes further. It reads the text inside your files — using optical character recognition for images and PDFs, native text extraction for documents and spreadsheets, and provider-native content APIs for Google Drive files. It then classifies what each document contains, extracts key details like dates, amounts, and person names, and builds a searchable understanding of every file in your vault.
When you search "electricity bill February", Astra doesn't just look for files with those words in the name. It looks for files whose content — the actual scanned text inside the image or PDF — contains electricity-bill-related content from February. It finds "IMG_20260215_scan.jpg" because it has read the scan and knows it's a utility bill.
Natural language in three languages
Astra understands queries in English, Hindi, and Hinglish — the mixed English-Hindi spoken by much of KeepCloud's user base. You can search for "mera train ticket dhundho" and Astra will correctly interpret this as a search for your train ticket. Noise words in all three languages — "dhundh ke batao", "please find me", "dikhao", "chahiye" — are automatically stripped so only the meaningful search terms reach the search engine.
Astra also auto-corrects typos by comparing your query against your own file vocabulary. If you type "elecricty bill" and have files related to electricity bills, Astra recognizes the likely correction and searches for the right thing.
"Astra understands: 'mera train ticket dhundho Jaipur wala' → search terms: ['train', 'ticket', 'jaipur']. And it finds the right IRCTC PDF, even if it's named 'scan_0042.pdf'."
The five-agent architecture
Astra is not a single function that runs a search. It's a coordinated system of five specialized agents — each responsible for one step in the pipeline:
Agent 1 — Query Preprocessor: Receives your raw query, strips noise words in all three languages, auto-corrects typos, and expands abbreviations (like expanding a city abbreviation to its full name). Outputs a clean set of search terms.
Agent 2 — Content Reader: Reads inside your files. Uses OCR for images and scanned PDFs, text extraction for Word/Excel/PowerPoint documents, and native content search APIs for cloud-based documents. Caches results to avoid re-reading the same file twice.
Agent 3 — Semantic Engine: Classifies documents by type (train tickets, invoices, ID cards, resumes, and 11 more categories), extracts named entities like dates, amounts, station codes, and person names, and builds a searchable tag cloud for each document.
Agent 4 — Verification Agent: Independently scores every candidate result from 0 to 100% confidence. Results below a minimum confidence threshold are silently discarded. This is what prevents false positives — files that superficially match your query but don't actually contain what you're looking for.
Agent 5 — Orchestrator: Coordinates the other four agents, runs search steps in parallel where possible, merges results from multiple sources, removes duplicates, and returns the final ranked list of files.
What AstraClaw AI does not do
Astra is not a large language model (LLM). It does not generate text, summarize documents, or answer questions about your files. It is a precision search engine — focused entirely on finding the right file fast, with high accuracy and low false-positive rates. This design choice means Astra is fast (results typically in 1–3 seconds), predictable (you can understand why a result was returned), and private (your file contents are processed locally on KeepCloud's servers, not sent to any third-party AI API).
Figure 1 — AstraClaw AI coordinates five specialized agents to understand your query and find the right file.
