Skip to main content
Gildea is a structured knowledge base of expert reasoning on AI. It tracks events and expert analysis from 500+ sources, decomposes each into independently verified text units, and connects them through shared entities and themes. Build on it directly via REST API or MCP, or let it power an agent that delivers the right insights at the right time.

​
Events + expert analysis

Gildea ingests both news events (what happened) and expert analysis from top investors, operators, researchers, and analysts (why it matters). Both are decomposed, verified, and linked through the same entity and theme system. NVIDIA announces H200 shipment numbers (event). Three days later, an analyst argues this shifts the infrastructure bottleneck from chips to power (expert analysis). Both are entity-linked to NVIDIA, tagged under Infrastructure, and decomposed into verified text units. You can trace from the event to the expert reasoning about what it means, programmatically.

​
Verified decomposition

Every signal is broken into its reasoning structure:
  • Expert analysis signals: thesis sentences, argument sentences, and claims
  • Event signals: summary sentences and claims
Each text unit is independently verified against source evidence. Only verdict=pass data is served.

​
Capabilities

Entity Intelligence

Companies, people, and products with trend analytics, significance testing, share of voice, theme distributions, and co-occurrence graphs.

Themes

Two-axis classification: 6 value chain segments and 6 market forces. Filter any endpoint by theme. Full reference.

Semantic Search

Hybrid semantic + keyword search across all verified text units. Search at the sentence and claim level, not just article titles.

Trend Analytics

Theil-Sen slope, significance testing (Mann-Kendall for entities, chi-square for themes), share of voice, volatility, and streak tracking.

​
Data shape

A single signal response contains the full verified decomposition tree: 
{
  "signal_id": "sig_01JABCDEF123456789",
  "title": "NVIDIA H200 Shipments Surge as Supply Eases",
  "content_type": "analysis",
  "tags": {
    "value_chain": ["Infrastructure"],
    "market_force": ["Competitive Dynamics"]
  },
  "entities": [
    {"entity_id": "org:/nvidia", "name": "NVIDIA", "type": "Organization"}
  ],
  "decomposition": {
    "thesis": {
      "text": "NVIDIA's H200 shipments increased significantly in Q4...",
      "sentences": [
        {
          "unit": {
            "unit_type": "thesis_sentence",
            "text": "NVIDIA's H200 shipments increased significantly in Q4."
          },
          "verification": {
            "final_verdict": "pass",
            "scoring_mode": "regression",
            "primary_score": 0.8912
          }
        }
      ]
    },
    "arguments": [
      {
        "sentences": [
          {
            "unit": {
              "unit_type": "argument_sentence",
              "text": "Supply chain bottlenecks eased following TSMC capacity expansion."
            },
            "verification": {
              "final_verdict": "pass",
              "scoring_mode": "regression",
              "primary_score": 0.8234
            }
          }
        ],
        "claims": [
          {
            "unit": {
              "unit_type": "analysis_claim",
              "text": "NVIDIA increased H200 shipments by 40% quarter-over-quarter."
            },
            "verification": {
              "final_verdict": "pass",
              "scoring_mode": "nli",
              "primary_score": 0.7421
            }
          }
        ]
      }
    ]
  }
}
Claims use NLI scoring and sentences use regression scoring. These are different methods. Do not compare primary_score values across scoring modes. Use the verdict as the primary trust signal.

​
Coverage

  • Sources: 500+ analysts and publications across the AI ecosystem
  • Data: January 2026 onward, updated daily
  • Source types: Research papers, SEC filings, podcasts, essays, news articles, social media, earnings transcripts, press releases
  • Themes: 6 value chain segments + 6 market forces (full reference)

​
Trust contract

Only verdict=pass data is served. Every text unit goes through:
  1. Evidence sourcing from the original article
  2. Factual consistency scoring with hard pass/fail thresholds (NLI for claims, regression for sentences)
  3. Deterministic safety checks catching contradictions, quantity mismatches, date errors, epistemic drift, and more
  4. Human review of flagged items with full audit trail
If a text unit fails at any stage, it is excluded from API responses. See Verification for full details.

​
Access

InterfaceBest for
REST APIApplications, agents, dashboards, data pipelines
MCP ServerAny MCP-compatible AI assistant or development environment
Slack AppTeam briefings and signal routing

Quickstart

First request in 5 minutes

API Reference

Full endpoint documentation