New · Zero-copy governanceTalk to founders
[ 200 OK ][ .SQL ][ MODEL ][ .JSON ]

Govern enterprise data with zero-copy virtualization

SOFI creates governed virtual views over private databases, APIs and files, so apps, BI and AI agents can use data without ETL or replication.It runs inside your VPC or on-premise.

// trusted access layer

Built for teams that cannot send raw enterprise data to every tool.

Private deployment

VPC / on-prem

Governance

RBAC + masking

Auditability

Every read logged

Agent surface

MCP + typed tools

[ 01 / 08 ]Main features
// Developer first //

Govern your data where it lives

The infrastructure layer that helps enterprise teams find, model, govern and publish data products.

// governance flow
Ready
1

Connect

postgres.core

2

Model

customer_360

3

Govern

pii.masked

4

Publish

sql + rest + ai

Lineage preserved from source to endpoint

Every published product keeps source, owner, policy and consumer context.

// schema preview

customer_id

uuid

key

email

varchar

masked

lifetime_value

decimal

visible

last_order_at

timestamp

visible

200 OK

// plug into your stack

Native bridges to AI agents and automation

// already integrated with

OpenAIAnthropicLangChainMCPn8nZapierCursor
[ 02 / 08 ]Agent ready
// AI access //

Drop SOFI into any AI stack

MCP-native - every governed view is also a typed tool. RBAC, masking and audit run on every agent call. No glue code, no leaking PII to LLMs.

mcp://sofi/acme/customer_360

// agent portal

Connect the client once

ClientClaude / Cursor
AuthSOFI_API_KEY
Scopeworkspace
// every call

Four guarantees, enforced before the agent sees a row

Same plane as REST and SQL — agents inherit the same policies, with no extra wiring.

  1. 1

    Identity

    JWT or API key resolves the agent role

  2. 2

    RBAC

    Policy match for purpose + scope + plan

  3. 3

    Mask

    PII redacted column-by-column before return

  4. 4

    Audit

    Agent ID, prompt fingerprint, row count logged

avg overhead<6 ms per call
[ 03 / 08 ]Core
// Engineered for prod //

Built for the workloads that won't tolerate slow

Predicate pushdown across 9 engines. DuckDB stitches the rest. Cache, policy and audit run in the same query path - not a sidecar.

<800ms

p95 federated query

9engines

with native pushdown

0bytes

leaving the source

100%

audit + lineage coverage

// benchmark

Same governed view, three execution paths

SELECT … FROM customer_360 WHERE region = 'EU' across Postgres + Mongo + Parquet, with RBAC and masking enabled.

ETL copy + app-side filtering4.2s
SOFI · source pushdown320ms
SOFI · policy-aware cache8ms
// p95 by view (last 5 min)streaming
customer_36082ms94ms
billing.invoices61ms71ms
events.last_7d118ms126ms
consent.subjects44ms49ms
agents.audit_tail38ms42ms
viewcoldcached

Reference workload: Postgres 16 (4 vCPU), Mongo 7 (Atlas M30), Parquet on S3. Numbers vary with selectivity, source latency and cache policy.

// why it's fast

Four engineering choices behind the numbers

  • Predicate pushdown

    Filters, joins, aggregates and CTEs rewritten by SQLGlot to the source's native dialect.

  • Parallel federation

    DuckDB stitches multi-source results in-memory. No intermediate landing zone.

  • Hot-path cache

    Per-view cache keyed by policy decision — never serves a stale row to the wrong role.

  • Same path as governance

    RBAC, masking and audit are in the query plan — not a wrapper that can be bypassed.

[ 04 / 08 ]Features
// Zero configuration //

We handle the hard governance

Masking, access control, lineage, auditability, caching and publication patterns in one layer.

SQL

// Sources to assets

Federate without copying

Files, operational databases, APIs, lakehouses and analytical stores become first-class governed assets — discovered, owned, lineaged.

****
****
****
name
role
dept

// Policy in the plan

Mask before the row leaves

RBAC, column masking and purpose-of-use are enforced inside the query plan — not a wrapper. No way to bypass.

CacheSource

// Live or cached

Policy-aware cache

Cache is keyed by the policy decision. A masked role never sees a cached unmasked row, ever.

subject:0x1f3asupportallowed
subject:0x9c44marketingdenied
subject:0xb201billingallowed

// LGPD by default

Consent-aware access

Per-subject opt-outs, retention windows and purpose checks resolved at query time — no quarterly spreadsheet reconciliation.

// audit.tail()live
agent=ai_support  view=customer_360
role=tier1        rows=14  mask=email,cpf
purpose=ticket    decision=allow
ts=2026-05-06T11:42:18Z

// Every read

Audit you can SQL

Every access — human, app or agent — produces a structured event with prompt fingerprint, role, scope and row count. Queryable like any other view.

SQL

REST

AI

Audit

Mask

Lineage

Preview

Revoke

// One surface

Publish, preview, revoke

Roll a governed view forward, snapshot a version, deprecate a column or revoke an agent — without re-deploying consumers.

[ 05 / 08 ]Use cases
// Use cases //

Transform data into governed products

See how SOFI turns private enterprise data into controlled, reusable access patterns.

// choose workflow

Customer 360 built without a warehouse copy

Join CRM, billing and product events into one governed product for support and revenue teams.

CRM

Salesforce

join: customer_id

Billing

Stripe

join: customer_id

Events

Kafka

join: customer_id

↓ JOIN customer_id
// governed viewlive

customer_360

One reusable surface for support, revenue and AI agents — RBAC + masking applied at query time.

// schema · customer_360

customer_id

uuid

key

email

varchar

masked

lifetime_value

decimal

visible

last_order_at

timestamp

visible

tier

enum

visible

cpf

varchar

masked

No warehouse copy

Pushdown across 3 live sources.

[ 06 / 08 ]Signals
// Design partner signals //

What early teams are validating

Patterns from design-partner conversations and pilots - not a public logo wall.

design signal
The strongest pull is replacing one brittle replication path with a governed virtual view that still reads from the source.

Design partner signal

Data platform team

design signal
AI access only makes sense if the same RBAC, masking and audit trail apply before the agent sees a row.

Design partner signal

AI platform team

design signal
The first useful milestone is not a migration project. It is one governed Customer 360 view across live systems.

Design partner signal

Enterprise architecture

design signal
LGPD work gets easier when lineage, policy decisions and access logs are queryable instead of tracked by spreadsheet.

Design partner signal

Governance lead

design signal
Private deployment is not a procurement checkbox. It is the reason production data can stay inside the existing boundary.

Design partner signal

Infrastructure team

design signal
Predicate pushdown, policy-aware cache and pgwire access need to feel like one query path, not three glued systems.

Design partner signal

Data engineering

[ 07 / 08 ]Comparison
// How SOFI compares //

Less operational weight, faster validation

SOFI is built to validate governed access quickly without standing up another replication-heavy data platform.

<2 wk

target for first governed view

guided design-partner rollout

0 copies

replication required for access

query sources through virtual views

1 layer

policy path for SQL, REST and MCP

no duplicated app-side filtering

// capability
SOFI
Legacy DV
DIY ETL
Zero-copy access across heterogeneous sources
Governed views for SQL, REST and agents
Partial
Custom
Policy-aware cache and pushdown in the query path
Limited
Self-hosted private deployment
Add-on
Built-in LGPD / RBAC / masking
Partial
Native AI agent surface (REST + MCP)
Custom
Time-to-production
< 2 weeks
3-6 months
6-12 months
License model
Per tenant
Per CPU core
Engineering hours

// why private deployment

“Customers want one governed layer across their stack — without lifting a single byte out of their environment. That's the deployment model SOFI was designed for.”

Enterprise positioning · 2026

0%

Data residency in private cloud

0+

Database engines federated

<0 wk

Time to first governed view

0%

Audit coverage on access

// need a performance benchmark?

Request a benchmark
[ MODEL ][ GOVERN ][ PUBLISH ][ AGENT ]

// get started

Ready to get started?

Spin up a governed view across your private data in under two weeks. No data leaves your environment — and every read is logged from day one.

Private deployment available · On-premise ready

[ 08 / 08 ]FAQ
// FAQ //

Frequently asked questions

Everything prospects need to understand the SOFI deployment model.

// 11 questions · 3 categories

Is SOFI still available as SaaS?

Only as a demo and starter environment. The strategic offer is the self-hosted PaaS / on-premise deployment.

What channels can consume SOFI views?

SQL users, REST/OData clients, internal applications, BI tools and AI agents can reuse the same governed views.

How long does a typical on-premise rollout take?

Most customers go from connector validation to first governed view in under two weeks of guided onboarding.

Which databases are supported out of the box?

PostgreSQL, Oracle, SQL Server, MySQL, MongoDB, ClickHouse, Cassandra, Neo4j, Elasticsearch, plus CSV, JSON, Parquet and Excel.