knowte 0.2.0

Local-first research discovery workspace combining academic databases and web search.

Knowte

Search wider. Read smarter.
Search academic databases and the open web from one local research workspace.

---

🚀 Spotlight

🔭 One search across research sources

Search arXiv, OpenAlex, Semantic Scholar, and an optional SearXNG web source. Knowte applies your area and year filters, merges the responses, and removes duplicate papers.

🧠 Intent-aware discovery

Connect any OpenAI-compatible cloud or local service. Intelligent Search expands academic retrieval terminology, ranks papers with Embeddings, and asks an LLM to verify the strongest academic and Web candidates against the full intent.

🖥️ A local workspace you control

Knowte runs on your machine. Configuration and usage data stay in ~/.knowte/, and the Python runtime uses only the standard library.

---

⚡ Quick Start

Knowte requires Python 3.9 or newer.

1. Install and start

python -m pip install knowte
knowte

Open http://127.0.0.1:7880 in your browser.

The default address is 127.0.0.1:7880. Specify a different address only when needed:

knowte --host 127.0.0.1 --port 8080

If you are running Knowte from a source checkout, install it in editable mode instead:

python -m pip install -e .

2. Choose your sources

Knowte enables all three academic backends by default, so you can search immediately. Open Config if you want to change the selected sources:

• arXiv works without additional configuration.
• OpenAlex works immediately; adding your email is recommended.
• Semantic Scholar accepts an optional API key.
• Web Search is off initially and requires SearXNG. See the next section.

Leave Max papers per search (N) at its default value unless you want a smaller or larger academic result set. Select Save after changing any Config option.

3. Run a search

Return to Search:

1. enter keywords, an author, or a research topic;
2. optionally choose one or more research areas;
3. optionally enter a start and end year;
4. choose Keyword or Intelligent, then select Search.

Keyword works without an AI service. Intelligent must first be configured as described below.

Each academic result provides whichever direct links are available: Paper, PDF, and DOI. Web results open their original pages.

Use Find More to continue the same search. Knowte reuses recently fetched academic candidates when possible and requests the next Web page only when needed.

---

🧠 Enable Intelligent Search

Knowte supports services that implement OpenAI-compatible /chat/completions and /embeddings endpoints. The service may be hosted by a cloud provider or run locally.

Open Config → Intelligent Search · OpenAI-compatible and fill:

• AI Base URL — the API root, normally ending in /v1, such as https://provider.example/v1 or http://127.0.0.1:11434/v1. Do not append /chat/completions.
• AI API key — the provider key. It may be blank when a local service does not require authentication.
• Language model — the provider's exact chat model ID.
• Embedding model — the provider's exact embedding model ID.

By default, both models use the same Base URL and API key. Enable Use a separate connection for Embeddings only when the embedding model is served elsewhere; then also fill Embedding Base URL and, when required, Embedding API key.

The result and verification controls balance coverage, cost, and latency:

• Keyword results also determines the academic candidate batch used by Intelligent Search. Each original or expanded query requests up to that value, capped at each provider's one-request maximum of 100.
• Intelligent results — target number of results that pass final LLM verification; default 20. Knowte verifies candidates in batches until it reaches the target or exhausts the candidate pool.
• AI timeout — timeout for each LLM or Embedding HTTP request; default 45 seconds.

Select Save, return to Search, and choose Intelligent.

For academic sources, Knowte generates up to two retrieval variants, retrieves broad candidates, ranks their titles and abstracts with Embeddings, and asks the LLM to verify the strongest candidates. Web results skip expansion and Embedding ranking and go from SearXNG recall directly to LLM verification. When an explicit research area is selected, it remains a hard academic filter. Without one, the LLM may infer useful retrieval terminology but does not silently save an area filter.

API keys are stored only in ~/.knowte/config.yml; they are not returned by the Config API or copied into Plans. The file is written with user-only permissions on systems that support them. Intelligent Search sends the query, candidate titles, abstracts or Web snippets to the configured AI service, so choose a provider appropriate for the material being searched.

If an AI stage fails, Knowte reports the degraded stage and falls back to the best available recall or Embedding order. Keyword Search remains independent of the AI configuration.

---

🌐 Enable Web Search

Web Search is powered by SearXNG, a separate open-source metasearch engine. Knowte can set up and operate a private local instance, but Docker must already be installed and running.

You need:

• a working Docker CLI;
• Docker Compose;
• a running Docker daemon.

Docker Desktop, Docker Engine, Colima, and compatible alternatives are supported. You do not need to keep the Docker Desktop window open.

Let Knowte set it up

Open Config → Local Web Search and select Set up.

Knowte will pull the official docker.io/searxng/searxng:latest image, create a local-only service, verify its search API, enable Web Search, and save the endpoint automatically.

Port 8888 is preferred. If it is unavailable, Knowte tries ports 8889 through 8898. The resulting endpoint is shown and saved in Config.

The remaining controls are:

• Start / Stop — control the local SearXNG container.
• Update — pull a newer image and safely recreate the service when needed.
• View logs — toggle recent setup and container output.
• Remove — remove the managed container and its configuration.

The downloaded image remains cached after Remove for faster setup next time. Select Delete cached image too if you also want Docker to remove that image when it is not used elsewhere.

Knowte limits managed container logs to about 30 MB. Image downloads time out after 10 minutes, and a failed first setup is rolled back.

Use an existing SearXNG instance

You can use your own local or remote SearXNG service instead. Follow the official SearXNG installation guide and enable JSON output in settings.yml:

search:
  formats:
    - html
    - json

Then enable Web Search, enter its search endpoint in Config, and select Save:

http://127.0.0.1:8888/search

To verify an endpoint:

curl --noproxy '*' -s \
  'http://127.0.0.1:8888/search?q=alpha&format=json' \
  | python -m json.tool

For custom Docker installations, set KNOWTE_DOCKER_BIN to the Docker executable. Knowte otherwise uses Docker from the system PATH and common platform-specific locations.

See the SearXNG Search API documentation for more details. SearXNG is distributed under its own license.

---

🔎 Search Sources and Results

Source	Coverage	Optional configuration
arXiv	Preprints and open research papers	None
OpenAlex	Broad scholarly metadata	Contact email
Semantic Scholar	Papers and citation metadata	API key
SearXNG	General web results	SearXNG endpoint

Result count

N controls the final target for academic-only and mixed searches. It defaults to 100, which matches the candidate batch Knowte can request from each academic provider in one retrieval round.

Web-only searches do not use N. They return the actual number of results in the requested SearXNG page, and Find More requests the next page without assuming a fixed page size.

Mixed-source balance

When academic and Web sources are searched together, Knowte initially allocates result slots as follows:

• three academic sources + Web: 3:3:3:1
• two academic sources + Web: 4:4:2
• one academic source + Web: 7:3
• academic sources without Web: equal shares

These ratios are targets, not rigid caps. After filtering and deduplication, available academic results fill shortages from another source.

Recently fetched candidates and Web pages remain in memory for five minutes. Serving results from this cache does not increase request counters; contacting a provider again does.

---

🛠️ Development

python -m pip install -e .
python -m unittest discover -s tests -v
python -m compileall -q knowte tests
node --check knowte/web/app.js

Node.js is only needed for the optional JavaScript syntax check.

---

🧭 Project Status

Knowte is currently an alpha research discovery tool. The longer-term direction is a broader research workspace with persistent notes and AI-assisted synthesis.

Not implemented yet:

• persistent notes, collections, or annotations;
• AI summarization or multi-paper synthesis;
• authentication or multi-user isolation.

Bug reports, ideas, and careful feedback are welcome.

---

📄 License

Knowte is released under the MIT License.