Saad Al-TamariThe Data Sync Platform pulls external systems (CRM, a knowledge base, meeting transcripts) into the shared 87-table schema. It is configured through a chat interface rather than code, and it runs as twelve agents, each with one job.
Plugins are rows, not code
Adding a data source means creating a configuration row, not deploying a code file. The seven core agents handle any REST API given the right config, which is why a new integration can be set up from a chat window in minutes. The chat agent can discover an API, propose mappings, create the configuration, trigger a sync, explain an error and roll back.
The numbers that made the call
The gateway is built for app traffic: single operations, permissions, rate limits. It has no bulk upsert. Routed through it, one 150-record sync becomes 395+ HTTP round trips, runs straight into the rate limit, and makes the gateway a single point of failure for bulk traffic it was never designed for.
Going around it without going dark
Bypassing the gateway did not mean bypassing accountability. The platform still writes audit entries and source-mapping rows, so the gateway’s view of the data stays complete. Anyone reading the logs sees the sync exactly as if it had gone through the front door.
Safety nets for bulk writes
The full pipeline runs with the writer switched off and produces a preview report before anything is committed.
Every sync is tagged. Inserts are deleted and updates reverted to their prior values from the audit log.
Knowing when not to route through your own abstraction matters more than building the abstraction.