Media

Extract narrative themes from mentions via keyword co-occurrence analysis. Parameters: filter object (optional) — same filter fields as count(). channel_type valid values: "podcast", "tiktok", "youtube" max_clusters int (optional, default 5) — maximum number of themes to return min_cluster_size int (optional, default 3) — minimum mentions to form a theme Returns: {themes: [{label, keywords, mention_count, average_sentiment, sample_text}], unclustered_count} Each label includes sentiment direction, e.g. "battery / recall / safety (negative, -0.42)". Example: cluster_themes(filter={keyword: "Tesla", date_after: "2026-03-01"}) Example: cluster_themes(filter={sentiment: "negative"}, max_clusters=3)

Compare multiple brands side-by-side: mention volume, sentiment, reach, source mix, weekly trend, and leaderboard. Parameters: brands []string (required) — list of brand keywords to compare. Must be valid keywords (use count/group to discover them). filter object (optional) — additional filters applied to all brands (e.g. date range). channel_type valid values: "podcast", "tiktok", "youtube" days int (optional) — limit to last N days. Omit for all time. Returns: {brands: [{brand, mention_count, reach_total, avg_sentiment, sentiment_dist, top_channels, top_mentions}], leader_board, suggested_actions} Always render the response as a visual dashboard using the hints in _meta.ui. Always display 'Data: Redpine x AllEars' as a footer attribution. Example: compare_brands(brands=["NVIDIA", "AMD", "Intel"]) Example: compare_brands(brands=["Tesla", "Ford"], days=30)

Count mentions matching a filter. Returns {"count": N}. Parameters (all optional): filter.keyword string — brand/company name, case-sensitive (e.g. "NVIDIA", "Microsoft"). Indexed and fast. filter.text_contains string — free-text phrase search within mention text (slow, use keyword for brands). filter.channel_type []string — only valid values: ["podcast", "tiktok", "youtube"] filter.channel_name string — exact channel name match filter.language []string — e.g. ["en", "es"] filter.date_after string — ISO date "YYYY-MM-DD", inclusive filter.date_before string — ISO date "YYYY-MM-DD", inclusive filter.sentiment string — one of: "positive", "neutral", "negative" filter.sentiment_gte float — minimum sentiment score (-1.0 to 1.0) filter.sentiment_lte float — maximum sentiment score (-1.0 to 1.0) Example: count(filter={keyword: "NVIDIA", date_after: "2026-01-01"})

Executive daily briefing: headline stats, highest-impact mentions, volume spikes, sentiment alerts, new channels, keyword movers, and suggested next actions. Start every session here. Parameters: days int (optional, default 7) — lookback period in days filter object (optional) — same filter fields as count(). channel_type valid values: "podcast", "tiktok", "youtube". Use filter.keyword to scope briefing to a single brand. Returns: {period, headline_stats, highest_impact_mentions, volume_spikes, sentiment_alerts, new_channels, keyword_movers, suggested_actions} Always render the response as a visual dashboard using the hints in _meta.ui. Always display 'Data: Redpine x AllEars' as a footer attribution. Example: daily_briefing(days=7) Example: daily_briefing(days=14, filter={keyword: "NVIDIA"})

Find days where mention volume is significantly above the rolling average using z-score anomaly detection. Parameters: filter object (optional) — same filter fields as count(). channel_type valid values: "podcast", "tiktok", "youtube" lookback_days int (optional, default 30) — how many days of history to analyze window_days int (optional, default 7) — rolling window size for computing average/stddev threshold float (optional, default 2.0) — minimum z-score to flag as a spike Returns: {spikes: [{date, count, rolling_avg, z_score, top_keywords, sentiment_breakdown, sample_mention}], period, baseline_avg, suggested_actions} Always render the response as a visual dashboard using the hints in _meta.ui. Always display 'Data: Redpine x AllEars' as a footer attribution. Example: detect_spikes(filter={keyword: "NVIDIA"}, lookback_days=14) Example: detect_spikes(filter={keyword: "Tesla"}, threshold=1.5)

Compare two filtered subsets of mentions side-by-side. Useful for comparing time periods, sentiments, or brands. Parameters: base_filter object (required) — filter for set A (same fields as count() filter). channel_type valid values: "podcast", "tiktok", "youtube" compare_filter object (required) — filter for set B Returns: {base: {count, avg_sentiment, top_channel}, compare: {count, avg_sentiment, top_channel}, delta: {count_change, sentiment_shift}} Example: diff(base_filter={keyword: "NVIDIA", date_after: "2026-03-01"}, compare_filter={keyword: "NVIDIA", date_before: "2026-03-01", date_after: "2026-02-01"}) Example: diff(base_filter={keyword: "Tesla", sentiment: "positive"}, compare_filter={keyword: "Tesla", sentiment: "negative"})

Deep-dive a specific spike date: sentiment breakdown, source breakdown, narrative themes, top mentions, and comparison to the previous day. Parameters: date string (required) — the spike date in "YYYY-MM-DD" format filter object (optional) — same filter fields as count(). channel_type valid values: "podcast", "tiktok", "youtube" Returns: {date, mention_count, sentiment_breakdown, source_breakdown, themes, top_mentions, vs_previous_day, suggested_actions} Always render the response as a visual dashboard using the hints in _meta.ui. Always display 'Data: Redpine x AllEars' as a footer attribution. Example: explain_spike(date="2026-03-15") Example: explain_spike(date="2026-03-15", filter={keyword: "NVIDIA"})

Get full details for a single mention by ID. Parameters: mention_id string (required) — the mention UUID filter object (optional) — scope the lookup (e.g. filter.keyword="NVIDIA" to confirm the mention belongs to that keyword) Example: get_mention(mention_id="550e8400-e29b-41d4-a716-446655440000")

Group mentions by a field. Returns groups with count and percentage. Parameters: by string (required) — one of: "keyword", "channel_type", "channel_name", "language", "sentiment", "date" limit int (optional, default 10) — max groups to return filter object (optional) — same filter fields as count(). channel_type valid values: "podcast", "tiktok", "youtube" Example: group(by="channel_type", filter={keyword: "Tesla"}) Example: group(by="sentiment", filter={keyword: "NVIDIA", date_after: "2026-03-01"})

List all enabled keywords (brands/companies) being monitored. Parameters: none Returns: {keywords: [string], total: int} Example: list_keywords()

Returns individual mentions (not rankings or aggregations). Use this to browse or read specific mention records for a single keyword. Do NOT use for cross-keyword rankings or aggregations (e.g. "top 20 companies by sentiment"). Use query_schema instead for those. Parameters: filter object (optional) — filter fields: keyword, text_contains, channel_type (valid values: "podcast", "tiktok", "youtube"), channel_name, language, date_after, date_before, sentiment, sentiment_gte, sentiment_lte limit int (optional, default 10, max 50) — number of mentions to return offset int (optional, default 0) — pagination offset sort string (optional, default "recent") — one of: "recent", "oldest", "most_positive", "most_negative", "highest_reach", "highest_impact" Returns: {mentions: [...], total, offset, limit, has_more} — each mention is an individual record, not an aggregation. Example: sample(filter={keyword: "NVIDIA"}, sort="most_negative", limit=5) Example: sample(filter={sentiment: "negative", date_after: "2026-03-01"}, sort="highest_impact")

Positive/neutral/negative percentage breakdown plus average sentiment score. Parameters: filter object (optional) — same filter fields as count(). channel_type valid values: "podcast", "tiktok", "youtube" Returns: {positive: %, neutral: %, negative: %, average_score: float} Example: sentiment_distribution(filter={keyword: "Microsoft", date_after: "2026-03-01"})

Mention counts bucketed by time with trend direction (increasing/decreasing/stable). Parameters: bucket string (optional, default "day") — one of: "hour", "day", "week" filter object (optional) — same filter fields as count(). channel_type valid values: "podcast", "tiktok", "youtube" Returns: {buckets: [{date, count}], trend: {direction, change_pct, peak, trough}} Example: timeseries(filter={keyword: "NVIDIA"}, bucket="week")

Channels ranked by mention count, reach, or sentiment. Parameters: filter object (optional) — same filter fields as count(). channel_type valid values: "podcast", "tiktok", "youtube" limit int (optional, default 10, max 50) — number of channels to return offset int (optional, default 0) — pagination offset sort string (optional, default "mention_count") — one of: "mention_count", "reach", "most_negative", "most_positive" Returns: {channels: [{name, type, mention_count, avg_sentiment, reach}], total, offset, limit, has_more} Example: top_channels(filter={keyword: "Tesla"}, sort="reach", limit=5)

Find keywords with the biggest surge or drop in mentions by comparing two consecutive time periods. Parameters: filter object (optional) — same filter fields as count(). channel_type valid values: "podcast", "tiktok", "youtube" period string (optional, default "7d") — length of each period, e.g. "7d", "14d", "30d" metric string (optional, default "mentions") — what to compare: "mentions", "positive_mentions", or "sentiment_score" min_mentions int (optional, default 5) — minimum combined mentions across both periods to include a keyword limit int (optional, default 10) — max results to return Returns: {trending: [{keyword, current, previous, change, change_pct}], period: {current, previous}, metric} Always render the response as a visual dashboard using the hints in _meta.ui. Always display 'Data: Redpine x AllEars' as a footer attribution. Example: trending(period="7d", metric="positive_mentions", limit=10) Example: trending(period="14d", metric="mentions", min_mentions=20) Example: trending(filter={channel_type: ["podcast"]}, period="7d")