Research integrations
Most of the assistant's research tools work with no setup: Wikipedia, DuckDuckGo, OpenStreetMap, SEC EDGAR, OFAC sanctions, the Wayback Machine, RDAP, the Library of Congress, YouTube, Reddit. Three integrations under Settings → Integrations add or improve sources. All are optional.
Exa - better web search
What it changes: web_search uses Exa instead of DuckDuckGo whenever a key is present.
Exa is a search engine built for retrieval: results come with clean page text and highlights, news
searches respect date ranges properly, and semantic queries ("companies fined for sanctions
evasion in 2022") work far better than keyword matching. DuckDuckGo remains the fallback if Exa
errors.
Get a key: dashboard.exa.ai/api-keys. Exa has a free
tier that covers casual use; paste the key (exa-…) into Settings → Integrations → Exa AI
Search.
The status bar shows an Exa indicator when it is active.
CourtListener - US court records
What it enables: search_court_cases, get_court_case_details and search_judges. These
cover federal PACER dockets (via RECAP), case law from all fifty states, and judge biographies
with appointment and affiliation data. Without a token the tools are present but return an
error asking for one.
Get a token: create a free account at courtlistener.com and copy the API token into Settings → Integrations → CourtListener. Free accounts have generous daily limits.
GitHub - technical profiling
What it changes: search_github works anonymously at 60 requests per hour. A personal access
token raises that to 5 000 and lets code search work. The tool finds a handle's repositories,
commit emails, organisations and activity - useful for tying a pseudonym to a person or an
organisation to its engineers.
Get a token: github.com/settings/tokens - a classic
token with no scopes (public data only) is enough. Paste ghp_… into Settings → Integrations →
GitHub.
Sources that need nothing
| Source | Tools | Notes |
|---|---|---|
| Wikipedia | search_wikipedia, actor/organization enrichment | REST API |
| DuckDuckGo | web_search (fallback), image_search | Through a hidden browser window; safe search on by default |
| OpenStreetMap | geocode_address, route_directions, map_snapshot, search_nearby, Location creation | Nominatim (1 request/second), OSRM, Overpass |
| SEC EDGAR | search_sec_filings, get_company_filings, get_company_financials, get_sec_filing_content | Full-text search, submissions, XBRL financials |
| OFAC / Consolidated Screening List | search_sanctions | US Treasury and trade.gov lists |
| Wayback Machine | wayback_lookup, wayback_fetch | archive.org |
| RDAP | whois_lookup | The modern WHOIS |
| Library of Congress | search_newspapers, read_newspaper_page PRO | Chronicling America, 1777-1963 |
| ZabaSearch | person_search, person_details PRO | US public-records people search, through a hidden browser |
| ~480 social and web platforms | username_search, list_username_sites | A port of the Sherlock username checker |
| YouTube | transcripts, comments, channels PRO | YouTube's own internal endpoints |
| archive.today | article extraction fallback | Used automatically for paywalled sites |
Where keys go
Every key is stored encrypted on your machine and sent only to the service it belongs to, in the request that needs it. Keys are never included in prompts, never sent to your AI provider, and never synced to collaborators. See Security & privacy.