Returns AI-coded ACLED-style conflict and political violence events derived from GDELT news clusters. Each event follows the ACLED methodology with a 3-level classification (disorder_type → event_type → sub_event_type), actor pair coding, geographic location, fatality count, and a free-text narrative. Events link back to their source media event clusters via cluster_id.
API key format: gdelt_sk_<64-hex-chars>. Generate in Dashboard → Settings → API Keys.
Lookback window in days from date. Default 1 = today only.
1 <= x <= 30Anchor/end date (YYYY-MM-DD). Defaults to today UTC.
"2026-07-03"
Number of events to return (max 100). Default 20.
1 <= x <= 100Pagination offset.
x >= 0Full English country name where the event occurred (e.g. 'Ukraine', 'Sudan', 'Israel', 'Gaza Strip').
"Ukraine"
State/province name. Use with country. Examples: 'Donetsk Oblast', 'Khartoum', 'Aleppo'.
"Donetsk Oblast"
Top-level ACLED disorder classification.
Political violence, Demonstrations, Strategic developments ACLED event type.
Battles, Protests, Riots, Explosions/Remote violence, Violence against civilians, Strategic developments ACLED sub-event type — most specific classification level.
Government regains territory, Non-state actor overtakes territory, Armed clash, Excessive force against protesters, Protest with intervention, Peaceful protest, Violent demonstration, Mob violence, Chemical weapon, Air/drone strike, Suicide bomb, Shelling/artillery/missile attack, Remote explosive/landmine/IED, Grenade, Sexual violence, Attack, Abduction/forced disappearance, Agreement, Arrests, Change to group/activity, Disrupted weapons use, Headquarters or base established, Looting/property destruction, Non-violent transfer of territory, Other ACLED normalized actor-pair label encoding who did what to whom. Format: 'ActorType1–ActorType2' (lower-numbered type first), or 'ActorType only' for one-sided events.
State forces only, State forces–State forces, State forces–Rebel groups, State forces–Political militias, State forces–Identity militias, State forces–Rioters, State forces–Protesters, State forces–Civilians, State forces–External/Other forces, Rebel groups only, Rebel groups–Rebel groups, Rebel groups–Political militias, Rebel groups–Identity militias, Rebel groups–Rioters, Rebel groups–Protesters, Rebel groups–Civilians, Rebel groups–External/Other forces, Political militias only, Political militias–Political militias, Political militias–Identity militias, Political militias–Rioters, Political militias–Protesters, Political militias–Civilians, Political militias–External/Other forces, Identity militias only, Identity militias–Identity militias, Identity militias–Rioters, Identity militias–Protesters, Identity militias–Civilians, Identity militias–External/Other forces, Rioters only, Rioters–Rioters, Rioters–Protesters, Rioters–Civilians, Rioters–External/Other forces, Protesters only, Protesters–Protesters, Protesters–Civilians, Protesters–External/Other forces, Civilians only, Civilians–Civilians, External/Other forces–Civilians, External/Other forces only, External/Other forces–External/Other forces If true, only return events where civilians were the primary target.
If true, only return events with reported fatalities > 0.
Natural-language semantic search query. When provided, results are ranked by embedding cosine similarity (text-embedding-3-small, 1536 dims) instead of chronological order. The query is embedded server-side and matched against pre-computed event note embeddings. Only events with a stored embedding (all non-sentinel events) are returned. Fully combinable with all other filters (country, event_type, disorder_type, date, etc.). Examples: 'airstrikes on civilian infrastructure', 'protest crackdown', 'rebel advance territorial control', 'military offensive artillery shelling'.
"airstrikes civilians residential areas"