Synquil turns scattered business data into one synced, queryable source that AI tools can read from directly, under access rules you set per person. There are five mechanical stages: Connect, Unify, Sync, Control, Serve.
You authorize read access to your existing tools — Google Sheets, Notion, HubSpot, Airtable, Shopify, QuickBooks, Stripe — through each provider's own standard OAuth flow. Your credentials for that tool (password, API key) are never shared with Synquil directly; Synquil only receives the OAuth token the provider issues, scoped to read access.
This matters because it means revoking access is always done at the source — disconnect Synquil from Google or HubSpot directly, and the access is gone immediately, the same way you would revoke any other connected app.
Using a high-capability AI system and curated context about your data's structure, Synquil designs an optimal unified schema that brings your scattered sources together into one consistent model — resolving naming differences, data types, and overlapping records automatically.
This matters because your sources were never designed to agree with each other. A "customer" in your CRM and a "customer" in your billing system rarely share a column name or an ID format. Unifying them once means every AI tool that queries the result inherits a single consistent model, instead of each tool reconciling the differences on its own, inconsistently.
Data is kept current on a recurring schedule, with frequency depending on your plan. Updates are deduplicated against existing records, so AI tools always see one accurate, current version of each record rather than duplicate or conflicting rows from repeated syncs.
This matters because a one-time import is a snapshot that goes stale the moment your source data changes. Recurring, deduplicated syncing is what keeps the unified schema an accurate reflection of your tools, not a copy frozen at connection time.
Access to the unified schema is granted per person, and denied by default. Each member of a workspace holds one or more roles, and a role names the sources, tables, and columns it permits; effective access is the union of a person's roles. A grant made at the source or table level automatically includes data added there later, while a grant narrowed to specific columns does not, so precision never widens on its own.
This matters because it is only possible over a synced copy. A layer that passes questions through to each source tool can grant or withhold that tool, but it cannot grant three columns of a spreadsheet — the unit it controls is the connection, not the data. Holding the data is what moves the boundary from the tool to the column.
The unified data is exposed through a single, secured endpoint using the Model Context Protocol (MCP), which AI tools — Claude, Cursor, ChatGPT, Windsurf — can query directly in plain English. Each person connects with their own credential, and the schema the endpoint describes back is already narrowed to what that person may see. Every query is read-only, is parsed and checked against the caller's grants before it runs, and is written to an append-only log whether it succeeds or is refused.
This matters because it gives every AI tool you connect the same access path to the same data, instead of each tool needing its own bespoke integration with each of your sources. Enforcing at the endpoint rather than in the AI tool is also what makes the boundary reliable: it holds regardless of which client asked, or what the model was told to do.