The Context Engine.
Datavor that gets smarter.

A persistent local SQLite store of everything Datavor learns about your data — relationships, business rules, transform recipes, error patterns, accepted suggestions. Every session starts smarter than the last. No competitor has this.

Four engines, one memory.

The Context Engine isn't a single store — it's four cooperating subsystems, each accumulating a different kind of knowledge.

How it compounds.

Three sessions across one month — same user, same database, same AI. Watch what Datavor learns and how each session starts smarter.

What's actually in the context.db.

The Context Engine is a single SQLite file. No proprietary format, no encryption layer, no service to call — just SQL tables you can sqlite3 into and read directly. Here's the schema, simplified.

-- Schemas Datavor has seen, fingerprinted for change detection
CREATE TABLE schemas (
  id TEXT PRIMARY KEY,           -- connection_id + table
  schema_json JSON,              -- columns, types, FKs
  fingerprint TEXT,              -- hash for diff detection
  first_seen TIMESTAMP,
  last_seen TIMESTAMP
);

-- Business rules — applied automatically by relevant tools
CREATE TABLE rules (
  id TEXT PRIMARY KEY,
  scope TEXT,                    -- 'global', 'database:X', 'table:X.Y'
  predicate TEXT,                -- SQL or DSL
  description TEXT,
  created_at TIMESTAMP,
  source TEXT                    -- 'user' or 'accepted_suggestion'
);

-- Named, versioned transform recipes
CREATE TABLE recipes (
  id TEXT PRIMARY KEY,
  name TEXT UNIQUE,
  version INTEGER,
  transforms_json JSON,
  tags TEXT,
  created_at TIMESTAMP
);

-- Errors with their fixes, for proactive recall
CREATE TABLE errors (
  id TEXT PRIMARY KEY,
  pattern_hash TEXT,             -- for similarity matching
  context_json JSON,             -- what was attempted
  error_message TEXT,
  resolution_json JSON,          -- what fixed it
  occurred_at TIMESTAMP
);

-- Suggestions surfaced to user, with their disposition
CREATE TABLE suggestions (
  id TEXT PRIMARY KEY,
  type TEXT,                     -- 'schema_change', 'sync_recovery', etc.
  payload_json JSON,
  status TEXT,                   -- 'pending', 'accepted', 'dismissed'
  created_at TIMESTAMP,
  resolved_at TIMESTAMP
);

Run sqlite3 ~/.datavor/context.db .schema on any Datavor install to see your actual tables — these and a few internal ones for indexing.

What the Context Engine stores. And what it doesn't.

The Context Engine is on your machine. It stays on your machine. Free tier is 100% local — nothing leaves. Pro sends a tiny daily license-validation heartbeat (aggregate counts, no content). Here's the strict line, drawn:

The 11 MCP tools that talk to the Context Engine.

The Context Engine is exposed entirely through MCP — your AI tool reads and writes it through these 11 tools. Full reference in the docs.

Why no other tool has this.

Competitors that have some of these features keep them in their cloud, behind their UI. None expose them as a flat SQLite file your AI can query as freely as it queries your databases. That gap is the Context Engine.

Start with fresh memory.
End the month with hindsight.

The Context Engine ships with every Datavor install — Free or Pro. No setup, no configuration. The file starts empty and fills itself.