Savvina AI turns natural language into precise SQL — instantly, privately, on your own infrastructure. Bring your own LLM. Pay no per-query fees. Ship answers in seconds.
No SQL required · No data leaves your servers · No surprise bills
Community edition · Non-commercial · BSL 1.1
savvina aiWatch how a business user goes from a plain-English question to a clean result — no SQL, no waiting.
Demo video coming soon
In the meantime, try the live interactive demo above
20+ data sources — databases, warehouses, and object storage
No data warehouse migration. No model training. Deploy on your own servers and start asking real questions the same afternoon.
PostgreSQL, MySQL, MongoDB, SQL Server, BigQuery, Snowflake, and 15+ more — including object storage (S3, GCS, MinIO, R2). Choose exactly which schemas and columns the AI may see.
Savvina reads your schema and auto-creates business-friendly names, descriptions, and value mappings. Refine in minutes.
Open the chat. Type naturally. Savvina translates, validates, and executes against your database — read-only, always.
Drill down, filter, compare. Every thumbs-up trains the example library — accuracy compounds with every query.
Every question flows through a deterministic, auditable pipeline. Nothing leaves your perimeter.
Eight chart types. Smart defaults from column shape. A full editor for axes, groupings, trend lines and filters — no notebooks, no dashboards to wire up.
Savvina inspects column types and picks the right chart — gauge for rates, KPI for single values, line for time series, bar for categories.
Sum, Count, Avg, Min, Max — pivot any column into series. Stack bars, connect nulls, drop nulls, all from one editor.
Overlay linear regression or a configurable trailing moving average (2–20 rows) on any time series.
Add row filters with =, ≠, >, <, contains. Multiple filters AND together and apply before aggregation.
Instant CSV & JSON downloads, server-rendered XLSX, and multi-section PDF reports with embedded chart images.
Log or linear Y scale, custom min/max, rotated X labels, currency / percent / decimal formats — fine-tuned for any dataset.
Thirteen capabilities engineered for accuracy, privacy, and the realities of regulated data.
Type a question. Savvina constructs a precise, validated query, runs it safely, and returns clean results — in seconds. Multi-turn refinement included.
Mark columns as off-limits and they are never sent to the LLM. PII auto-detection. Query logs store metadata only — never values.
Native adapters for PostgreSQL, MySQL, MongoDB, SQL Server, Azure SQL, BigQuery, Snowflake, ClickHouse, Databricks, Trino, Athena, and 8 object-storage sources (S3, GCS, Azure Blob, MinIO, R2, and more).
11+ providers — Anthropic, OpenAI, Google, Groq, Cerebras, Mistral, Ollama, GitHub Models, HuggingFace, Together.ai, OpenRouter, and any OpenAI-compatible endpoint. Switch per session.
Auto-generated business names, descriptions, value mappings, default filters, reusable metrics, and join hints. Fully editable.
Exact-match plus semantic-similarity caching cuts LLM calls by 40–60% in production — slashing latency and inference cost.
Every thumbs-up is added to a few-shot library. Accuracy compounds with use — no manual tuning required.
Auto-execute for trusted users, Review-first for compliance, Generate-only for high-security DBAs. Configurable per connection.
One-command Docker Compose deployment. Runs on-prem, in your VPC, AWS, Azure, GCP, or bare metal. No external dependencies.
Admin / User roles. Per-user database credentials. Audit log with CSV export and configurable retention.
SAML 2.0, Microsoft Entra ID, SCIM provisioning. Offline licenses for air-gapped deployments. White-label available.
Select results from chat history and assemble a PDF report. Export individual query results as CSV, XLSX, or PNG.
Generate a read-only share link for any chat session. Send stakeholders a live view of results — no account required.
Most "AI for data" tools demand that your schema, query logs, and sometimes your raw rows traverse somebody else's cloud. Savvina was built the other way around: self-hosted, BYOLLM, with column-level controls baked into the engine — not bolted on later.
Mark any column off-limits — it is never sent to the LLM. Sensitive PII patterns are auto-flagged.
Every query is parsed and validated for safety before it touches your database. Writes are physically impossible.
Audit who ran what, when, and from which model — without ever logging actual data values.
Database credentials and LLM API keys are Fernet-encrypted before storage. Keys never appear in logs or backups.
Run fully offline with Ollama. Offline-signed licenses for environments with zero internet egress.
Whether you're answering board questions or building internal tooling, Savvina removes the bottleneck between people and the data they already own.
Sales, ops, finance, and PMs ask their own questions in chat — instead of waiting days for an analyst to write a one-off query.
Push routine lookups and aggregations to Savvina. Free your analysts for modelling, experimentation, and strategic work.
Healthcare, finance, legal — sectors that cannot send production data to third-party clouds. Self-hosted with column-level privacy.
Give internal teams safe database access in minutes. Docker deploy, no ongoing engineering maintenance, no custom UI to maintain.
KPI checks — revenue, churn, pipeline — in seconds, from the actual source of truth. No analyst on standby, no week-old PDF.
Run actuals vs. budget by cost center, segment, or entity. Sensitive financial data never leaves your infrastructure — no third-party cloud sees your numbers.
Ship faster when you own your own questions. Stop blocking the data team for SQL you could write in plain English.
No SQL, no dashboards, just answers. High-frequency questions get instant answers without yet another Looker tile.
Mark salary, SSN, DOB, and address columns as never-leave-the-database. The LLM sees aggregates only. Payroll data stays locked.
/ And that's just the start
Beyond the personas above, any role that regularly waits on answers — supply chain planners tracking inventory turns, e-commerce teams chasing conversion by cohort, hospitality managers reading occupancy trends, research teams auditing experiment results, procurement leads monitoring vendor spend — can ask Savvina directly. If there's a database behind the question, there's no reason a human should be the bottleneck between it and the answer.
Flat per-workspace pricing with no query metering. You bring the LLM you already pay for.
Join the waitlist for early access to the commercial production release, or get in touch if you need more information.
No spam. Unsubscribe at any time.
Self-hosted production edition (commercial) coming soon. Cloud-hosted edition coming later.