A semantic model your team actually reviews
Physical tables become entities, attributes and metrics with business names and confidence scores. Edit anything; your overrides survive every rescan.
SchemaPilot discovers your schema, builds a reviewed semantic model, generates scored query tools your policy approves, runs reports with charts, watches for drift, and lets you deploy a Co-Pilot into your own apps. Every answer comes from an approved tool.
AI may propose, interpret, explain and select. AI may not authorize, publish or execute ungoverned data access.
The same lifecycle for every source, no application-specific code. Each stage leaves a reviewable artifact: a snapshot, a model, a tool revision, an audit entry.
Register a database with a read-only account, or paste a connection URL. Credentials are encrypted at rest with AES-256-GCM and edited or removed in place.
SchemaPilot profiles tables, keys, indexes and routines, proposes a semantic model with confidence scores, and lets your LLM enrich names, descriptions and metrics as a reviewable proposal.
Schema analysis writes KPI, trend, breakdown, ranking and record tools with a 0–100 usefulness score, risk level and rationale. Your governance policy auto-approves the safe ones; you review the rest.
Run tools with charts and tables, ask the Co-Pilot in plain language, and embed a deployed Co-Pilot in your own apps behind API keys. Raw SQL is never generated.
Physical tables become entities, attributes and metrics with business names and confidence scores. Edit anything; your overrides survive every rescan.
Tools persist normalized query plans validated against the catalog. Approval is a human action with a revision number.
Rescans stage immutable snapshots. Renames, drops and type changes list the exact tools they affect, and only those are suspended.
Ollama is used by default. Save your OpenAI key to opt into the hosted provider; the key is encrypted with your organization id bound as associated data and decrypted only in memory for a single request.
Schema analysis decides what is worth asking: fact tables get KPIs and trends, lookups get lists, joins become labelled breakdowns. Each tool shows its score, risk and rationale, and your governance policy approves the safe ones automatically.
One script tag or a REST call. Publishable keys for browsers, secret keys for servers, origin allow-lists, rate limits and charts in the answers. Governed exactly like the app.
Tables, views, columns, foreign keys, indexes, routines, comments and bounded sample profiles from every supported engine, as immutable catalog snapshots.
Entities, attributes and metrics with business names and confidence scores. Your edits survive rescans; LLM enrichment arrives as a stored proposal you accept per entity or discard.
Table roles, column roles, sensitivity, join paths and soft-delete filters decide which tools exist. Each carries a score, a risk level, a category and the reason it was written.
Set a minimum score, maximum risk and allowed categories once. Safe aggregate tools are approved at scan time; everything else waits for a person. Apply a new policy to existing sources in one step.
Proposed, approved, published, suspended, retired, reactivated. Optimistic revisions mean only the exact version your team reviewed can run. AI-suggested tools are always Proposed first.
Rescans stage snapshots and list the exact tools a rename, drop or type change affects. Breaking changes suspend only those tools until the review is accepted.
Run any approved tool with validated parameters. Results open as KPI tiles, trends or breakdowns with a one-click table view, in the app and inside deployed Co-Pilots.
Embed a chat widget with one script tag or call the REST API with hashed keys. Allowed origins, per-key rate limits, suggested questions and a monthly execution quota keep it governed.
Long LLM runs execute on the server with live progress. Leave the page, come back, and the result is waiting. Cancel at any time.
Every execution, Co-Pilot conversation and lifecycle move is logged with typed arguments and a sanitized outcome. Never credentials, never result rows.
Ollama is used by default. Save an OpenAI key to opt into the hosted provider for your workspace; it is encrypted per tenant and removable at any time.
Six-digit codes for email verification and password reset, every other session signed out on a password change, and card-backed trials through Stripe.
Execution authority belongs to deterministic policy and approved, versioned tool contracts. The language model is a planner, never an executor.
Illustrative example of a compiled preview; your tools are generated from your own schema.
Every plan starts with a 7-day free trial. Add a card, cancel before the trial ends and you pay nothing. Bring your own LLM key; we never resell model usage.
For a single team bringing its first database under governance.
For data teams running governed analytics across several systems.
For organizations standardizing AI access to enterprise data.
Need more than 50 databases, SSO, or a private deployment? Compare every plan.
Yes. Deploy a Co-Pilot bound to one data source, restrict it to selected approved tools, allow the origins it may run on, and embed it with a single script tag or call the REST API with a secret key. Every answer still goes through the same approvals, quota and audit trail as in the app.
Checking for changes stages a new snapshot and lists every tool a rename, drop or type change would break. Only those tools are suspended, until you accept the review (which migrates the semantic model and regenerates tools) or dismiss it.
No. Questions are routed to tools that your team has already approved. Each tool is a normalized query plan validated against the discovered catalog; the SQL you see is a compiled preview, never the execution authority.
PostgreSQL, Microsoft SQL Server and MySQL / MariaDB today. The connector layer is pluggable, so additional engines can be added without changing the governance model.
Ollama is used by default. You can explicitly save your OpenAI key in Settings to use OpenAI for your workspace. The key is encrypted with a per-tenant binding and decrypted only in memory for each request.
Only tool names, descriptions and parameter names for routing, the parameter contract for argument extraction, and at most 20 aggregate rows for answer wording. Never schema, SQL, credentials or row-level data.
Yes. Pick any plan and start a 7-day free trial. A card is required so the plan can begin automatically when the trial ends; cancel before then and nothing is charged. The Co-Pilot unlocks once the paid plan starts. Billing is handled by Stripe.
7 days free on any plan. Cancel before the trial ends and you pay nothing.