Private metasearch. Keyless indexes, AI on your own key, history that never leaves this browser.
j / k move · Enter open · / focus search
AI mode
The current query and the top retrieved results are sent as numbered context to your provider. The answer cites those result URLs below it. The key is stored only in this browser and sent browser → provider; seearch never receives, stores or pays for it.
AI provider (BYOK)
OpenAI-compatible base URL, model and key. Examples:
https://api.openai.com/v1, https://openrouter.ai/api/v1,
or a local http://localhost:11434/v1.
History is local only: this browser's localStorage. The server never receives your key or your AI history.
Similarity check (upload)
Upload .txt, .md, .csv or
.pdf. The file is split into chunks and compared with the current
result snippets and the local web index. The output is a
similarity signal, not a verdict — quotations and missing
sources affect it. Scanned PDFs with no text layer are reported as
no extractable text (OCR is not performed). Files are processed
transiently and are not stored.
History is local only. Search history and AI
history are stored separately in this browser's localStorage. They
are never sent to the server and can be deleted here at any time.
Search history
No searches yet.
AI history
No AI answers yet.
What this is
Keyless metasearch over DuckDuckGo plus the estate's own corpora, a BYOK
AI mode that runs on your key, a transient similarity check and
timestamped report exports. Nothing is stored on our servers: API responses
are no-store, and your search and AI history live only in this
browser's localStorage.
Search has a hard latency budget (1.9 s corpus fan-out, 2.6 s whole request). If a corpus misses the deadline the page says Partial results instead of pretending the index is complete. Each alternative backend (Marginalia, Mojeek, Stract, Wikipedia, OpenAlex, arXiv, PubMed, Gutenberg, and the go-v corpus) has its own 1.2 s deadline; an unreachable engine is shown as unavailable with its measured latency.
Reports
Every report includes the query, each source URL, the engine and a UTC
retrieval timestamp. Formats: Markdown, CSV, JSON, standalone HTML, DOCX and
XLSX. The DOCX/XLSX files are minimal OOXML zip containers built by the
Python standard library. Downloads are named
seearch-report-<query-slug>-<YYYYMMDD-HHMMSS>.<ext>
with the timestamp in UTC.
Indexing
The service publishes /sitemap.xml and pushes changed pages with IndexNow to Bing, Yandex, Seznam and Naver. No paid APIs are used. Google does not participate in IndexNow: verify the site in Google Search Console and submit the sitemap there. Details: INDEXING.md.
Similarity, not a verdict
The upload check compares word sequences against retrieved snippets and the local index. It cannot prove authorship or plagiarism; common phrasing, quotations and missing sources all affect the signal. Always review the matched passages yourself.