findora 2025.5.11803

Search engine

findora

findora is an LLM-assisted search utility.
Give it a plain-text query and it returns a curated list of links with titles and descriptions, after:

• validating the query with a small language-model prompt
• optionally enhancing that query for better recall/precision
• running iterative searches while:
  • deduplicating URLs
  • honouring language & geo preferences
  • respecting result and iteration caps

It is ideal for chat-bots, data-gathering pipelines, academic tooling, or anywhere you need a “few good links” rather than an HTML page of unstructured results.

---

Installation

pip install findora

Python ≥ 3.8 is required.

---

Quick start

from findora import findora

results = findora(
    search_query="Eugene Evstafev",
    n=10,
    enhance=False,
    verbose=False,
    language="en-US",
    location="UK",
)

print(results)

Example output (truncated/pretty-printed):

[
    {
        "title": "Eugene Evstafev – Google Scholar",
        "url": "https://scholar.google.com/citations?user=cYLfW7QAAAAJ&hl=en",
        "desc": "Eugene Evstafev, University of Cambridge. Topics include computer science …",
    },
    {
        "title": "How A24 Changed Contemporary Cinema – Medium",
        "url": "https://medium.com/@chigwel/how-a24-changed-contemporary-cinema-5dc69c0b00c2",
        "desc": "Article discussing the impact of A24 films and the 2023 Oscars success …",
    },
    …
]

---

API

findora(
    search_query: str,
    llm: Optional[Any] = None,   # custom ChatLLM7-compatible object
    n: int = 10,                 # max results (1 – 10)
    enhance: bool = True,        # turn query rewriting on/off
    verbose: bool = False,       # print prompt/response trace
    max_retries: int = 15,       # LLM retry budget
    language: str = "en-US",     # BCP-47 tag passed to the search prompt
    location: str = "World",     # free-text geographical hint
    max_iterations: int = 55,    # safety stop for the search loop
) -> list[dict]

Raises ValueError for:

• empty or > 1024-char queries
• n > 10
• invalid query as judged by the model.

---

Features

• 🔍 LLM-powered query understanding – prevents garbage queries early.
• ✨ Optional query enhancement – adds synonyms, context and localised terms.
• 🗺 Language & region control – pass language/location and let the model obey.
• 🚫 Deduplication & “exclude” list – zero duplicated links across iterations.
• 🧪 Deterministic unit-tested core – 100 % tests pass with mocked LLMs.

---

Contributing

Bug reports, pull requests and feature ideas are welcome!
Head over to the issue tracker to get started.

1. Fork the repository
2. Create a feature branch
3. Run the test suite with pytest
4. Open a PR

Please run black and isort before submitting.

---

License

findora is distributed under the MIT License.

---

Links

• 🐙 GitHub  →  https://github.com/chigwell/findora
• 📦 PyPI      →  https://pypi.org/project/findora/
• 👤 Author   →  Eugene Evstafev