{
  "id": 14516755,
  "title": "I built a Read.ai connector for Cognee. Importing once wasn’t enough",
  "url": "https://urgent.news/2026/10/14/i-built-a-read-ai-connector-for-cognee-importing-once-wasnt-enough",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-14T19:13:21.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/anamasgard/i-built-a-readai-connector-for-cognee-importing-once-wasnt-enough-3hcj"
  },
  "original_language": "en",
  "account": "A developer built a Read.ai connector for Cognee. Importing meeting reports once was not sufficient because those reports continuously change. The connector allows users to choose meeting IDs or date ranges, and specifies whether to include summaries, transcripts, and metrics for each report. The imported reports are stored as separate documents, preserving context such as participant names, timestamps, and the original report link.\n\nThe connector handles the workflow through the sync_read_ai() function, which manages ingestion, recovery, Cognee processing, and cleanup verification. Users can search the selected dataset after the import. The connector uses OAuth 2.1 for authentication, and handles token refresh to prevent concurrent runs from refreshing the same credential simultaneously.\n\nUnlike other synchronization methods, this connector rechecks selected reports, renders their content consistently, and compares hashes. This ensures that only new or changed content is sent downstream, avoiding repeated extraction and embedding work for unchanged documents. However, this method still requires re-reading the selected reports from Read.ai.\n\nThe connector successfully imported two fictional reports, a launch plan and a support handover, which became searchable documents. When the launch report was changed from Friday to Monday, the document's references and chunk vectors were updated, while the old document's references were removed. The removed report and its chunks were deleted from the storage.\n\nThe connector also demonstrated handling errors, such as a simulated 403 permission failure, which left the sync incomplete and preserved the previous corpus and successful checkpoint. The implementation flow includes five phases: updates, reuse, removal, and failure handling, but does not cover the quality of a real model's answers.\n\nWhile the recent local verification showed no failures or errors, the connector's implementation still requires work, such as verifying live Read.ai OAuth, handling field availability, and addressing deletion and permission transitions. The successful checkpoint must occur after ingestion, Cognee's processing step, and verification.",
  "summary": "Built for WeMakeDevs × Cognee Mergetober. A meeting report says the launch is on Friday. Someone corrects it to Monday. Your assistant still retrieves Friday. Nothing crashed. The import worked. The copy is just out of date. That is the problem behind the Read.ai connector I built for Cognee: importing meeting reports is useful, but those reports do not stop changing once they reach your…",
  "key_points": [
    "Developer created Read.ai connector for Cognee",
    "Connector allows selecting meeting IDs or date ranges",
    "Imported reports stored as separate documents with context"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}