[ilya-bogin-keenable]

Summary

Adds Keenable as an optional web search provider in backend_api_python/app/services/search.py, next to GDELT and DuckDuckGo as a free option. It needs no API key: set SEARCH_PROVIDER=keenable (it is added to the setting's options in routes/settings.py) and AI analysis and the research agent get web results with page text snippets and publish dates. An optional KEENABLE_API_KEY only raises the rate limits. Tavily stays the default and the existing fallback order is unchanged.

Related issue

• None. No new runtime dependency (uses requests like the other REST providers), no route, schema or migration change.

Changes

• KeenableSearchProvider(BaseSearchProvider): POST https://api.keenable.ai/v1/search/public without a key, POST https://api.keenable.ai/v1/search with X-API-Key when a key is set. Results map title, url, snippet (whitespace collapsed, capped at 500 characters like Tavily/SerpAPI/SearXNG) and published_at. days <= 30 becomes published_after (Google CSE applies dateRestrict up to the same 30-day cutoff); longer windows such as search_research send no date filter. HTTP errors and empty results return success=False, so the chain moves on to the next provider.
• SearchService._init_providers: Keenable is registered only when SEARCH_PROVIDER=keenable or KEENABLE_API_KEY is set. When selected it goes first through the existing preferred logic; with only a key it is appended before DuckDuckGo. No other provider moves.
• provider_status() and _search_provider_note(): a Keenable row and a setup note.
• routes/settings.py: keenable in the SEARCH_PROVIDER options and an optional KEENABLE_API_KEY password field (Advanced tab, next to SERPAPI_KEYS).
• utils/config_loader.py, .env.example, backend_api_python/env.example: KEENABLE_API_KEY mapping and an empty entry.
• tests/test_keenable_search.py: 7 offline tests (keyless and keyed request shape, snippet mapping and cap, date window, error and empty fall-through, not registered by default, selected provider first, key-only placement).

What is sent to Keenable: the search query, max_results, and published_after for short windows, with an X-Keenable-Title: QuantDinger header (required by the keyless endpoint), plus the key only if one is set.

Test plan

• Tested locally with docker compose up -d --build (not run; the search service was exercised directly, see below)

• Backend logs show no errors (server not started)

• Relevant pytest tests pass

• python -m pytest -m "not integration and not stress" on Python 3.12 (requirements.lock + requirements-dev.txt), macOS: 3688 passed, 286 failed. The same 286 fail on main here (3681 passed); 233 of them are strategy sandbox memory limit could not be applied, which macOS does not support. The 7 new tests pass.

• ruff check app scripts tests: passed.

• Live run without a key through get_search_service() with SEARCH_PROVIDER=keenable (provider order Keenable -> GDELT -> DuckDuckGo): search_stock_news for NVDA and BTC, search_with_fallback (7 days), search_stock_events (30 days) and search_research each returned 5 results from Keenable in 0.4 to 1.9 s. Every result of the four news calls carried a published_at inside the requested window.

API documentation (if routes/schemas changed)

• Not applicable: no route or OpenAPI schema change.

Disclosure: I work at Keenable, which runs this API.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues
0 New dependency risks

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud