in the language your stack already speaks — AI, PostgreSQL, or JDBC
AI agents, any PostgreSQL client, or any JVM tool — pick the surface that fits.
ofrag for Claude/Gemini · ofpgproxy for pgAdmin/DBeaver/Metabase/DuckDB/dbt · OFJDBC for the JVM
MCP Server for Claude, Gemini, and other LLMs. Semantic search across 25,000+ tables.
JDBC DriverJVM-native driver for DBeaver, Apache Airflow, Java/Kotlin/Scala data pipelines.
PostgreSQL WireNative PG endpoint — pgAdmin, DBeaver, Metabase, DuckDB, dbt, psql, asyncpg, pgx all just work.
An MCP Server that gives Claude, Gemini, and other LLMs structured, safe access to your Oracle Fusion data
Your AI agent receives a question like "Show me unpaid invoices from Acme Corp last month"
ofrag finds the right tables and REST APIs using vector embeddings and business process mapping — no hardcoded queries needed
SQL is parsed into an AST and validated against a cached schema. Only clean, verified queries reach your database
The validated query runs against Oracle Fusion and results come back in about 4 seconds, formatted and ready to use
PAYMENT_STATUS_FLAG = 'N', discovers the correct AP tables via semantic search, validates the SQL against the local schema cache, executes it on Oracle Fusion, and returns a formatted table — all without the user knowing a single table name.
Everything your AI agent needs to understand and interact with Oracle Fusion Cloud
Find relevant tables and APIs using natural language. Vector embeddings across 25,000+ table descriptions and 500+ REST resources. Understands business intent — "money from customers" finds AR tables even without exact matches.
Every query is parsed into an Abstract Syntax Tree and validated against a locally cached schema before execution. The AI gets precise feedback like "column VENDOR_NAME does not exist in AP_INVOICES_ALL" — preventing hallucinations.
Native integration with Claude, Gemini, and any MCP-compatible LLM. Tools include semantic_search, execute_oracle_sql, rest_call, lint_sql, module_summary, and more — giving AI structured access to your ERP.
Full CRUD operations on Oracle Fusion REST APIs. Self-discovering endpoints — the AI calls /describe to understand resource structure dynamically. Automatic pagination for bulk exports.
All metadata exploration happens locally — zero queries to production during discovery. Schema cache, fuzzy search, synonym expansion, and business process cross-referencing run at zero latency.
Switch between dev, SIT, UAT, and production instantly. Mixed authentication per environment (SSO or basic auth). Compare data across environments without configuration changes.
Free, open-source JDBC driver for the JVM ecosystem — DBeaver, Apache Airflow, Java/Kotlin/Scala/Groovy data pipelines.
Execute SQL queries directly against Oracle Fusion Cloud without creating BIP reports. Get instant results for troubleshooting, analysis, and data exploration with DBeaver's intelligent autocomplete.
Build complete ETL pipelines from Oracle Fusion to AWS (S3, Athena, QuickSight), Google Cloud (BigQuery, Cloud Storage), Azure (Synapse, Blob Storage, Power BI), or OCI (Object Storage, Data Lake).
Seamlessly integrates with DBeaver, the premier free database IDE. Local metadata caching for instant autocomplete, real-time error detection, and intelligent table/column suggestions.
Export query results to Parquet (partitioned), CSV, JSON, Excel, or load directly into DuckDB, PostgreSQL, MySQL, or cloud databases. Build data lakes without custom scripts.
Standard JDBC driver — works in Java, Kotlin, Scala, Groovy, Apache Airflow (JdbcOperator), Talend, Pentaho, and Apache NiFi. For non-JVM languages, use OFArrow with Arrow Flight SQL or ofpgproxy with native Postgres bindings.
Connect QuickSight, Power BI, Tableau, or Looker to Oracle Fusion data via cloud data warehouses populated by OFJDBC pipelines.
OFJDBC & DBeaver:
OFArrow (Arrow Flight SQL):
ofpgproxy — PostgreSQL Wire:
AI & RAG:
Any tool that speaks PostgreSQL connects to your Oracle Fusion tenant. No driver install, no BIP report, no custom adapter — just a connection string.
Any PG client — pgAdmin, DuckDB, dbt, asyncpg — opens a normal Postgres connection on port 5433
18+ translator passes rewrite Postgres idioms (LIMIT, ILIKE, regex, date_trunc, EXCEPT, recursive CTE) into Oracle SQL
Schema-introspection queries are answered from an embedded DuckDB metadata cache — never reach Oracle, never block the user
SOAP XML response is parsed in constant memory and streamed to the client row by row, with proper PG types (numeric, timestamp, bool)
127.0.0.1:5433. The dashboard's SQL uses native Postgres syntax — WHERE invoice_date >= NOW() - INTERVAL '1 month' AND payment_status_flag = 'N'. ofpgproxy rewrites NOW() to SYSDATE, the date math to Oracle's ADD_MONTHS semantics, the boolean predicate to 'N', and ships the request to BI Publisher. Results stream back into the chart. No BIP report. No ETL.
Built for production: streaming, cancellation, transaction state, type fidelity. Tested against 6 client toolchains in real Fusion tenants.
Implements the actual PG wire protocol, not a REST shim. Every PG client driver works unmodified — pgx (Go), psycopg / asyncpg / SQLAlchemy (Python), pgJDBC (Java), node-postgres, libpq, ADO.NET. No vendor lock-in.
LIMIT n → FETCH FIRST n ROWS ONLY, ILIKE → UPPER() LIKE UPPER(), ~ / !~ regex → REGEXP_LIKE, date_trunc → TRUNC, EXCEPT → MINUS, WITH RECURSIVE preserved, boolean TRUE/FALSE → 1/0 or Y/N depending on column type. Idioms translate transparently.
Schema-introspection queries (the bulk of what BI tools fire on connect) are answered from a local DuckDB-backed catalog. ~30,000 Fusion tables indexed offline. Connection setup is instant; production never sees a discovery probe.
BI Publisher's SOAP responses are parsed in constant memory regardless of result size. Million-row SELECTs flow through the proxy without OOM. Streaming is respected end-to-end — rows reach the client as they arrive.
DECLARE / FETCH / CLOSE for postgres_fdw and dblink. COPY (...) TO STDOUT text format for DuckDB's postgres_scanner and pg_dump. CancelRequest support — the Stop button in IDEs actually cancels the in-flight SOAP call.
BI Publisher cannot write, and ofpgproxy will never write. DML paths simply don't exist in the codebase — safe to point production users at without permission gymnastics. SSO authentication via Chrome with auto-reopen on token expiry.
ofrag is a Retrieval-Augmented Generation engine and MCP Server that enables AI agents like Claude and Gemini to intelligently query Oracle Fusion Cloud. It provides semantic search across 25,000+ table descriptions, AST-based SQL validation, and live query execution — letting your AI understand and retrieve ERP data using natural language.
Yes. ofpgproxy exposes Oracle Fusion as a native PostgreSQL endpoint. Point pgAdmin, DBeaver, DataGrip, or any other PG client at the proxy's local port (default 5433) and use it like any Postgres database. SQL is translated on the fly into BI Publisher calls — no BIP report creation required.
ofpgproxy makes Fusion look like Postgres. DuckDB connects via ATTACH (TYPE POSTGRES). dbt uses the dbt-postgres adapter. Metabase, Superset, and Grafana add a PostgreSQL data source pointing at the proxy. All standard SQL works — JOINs, GROUP BY, aggregates, ILIKE, regex, date_trunc, EXCEPT, recursive CTEs — translated transparently.
ofpgproxy is a read-only PostgreSQL wire-protocol proxy — any tool that speaks Postgres connects to it. OFJDBC is a JDBC driver — only JVM-based tools (DBeaver, Airflow, Java/Kotlin/Scala apps) can use it. Pick the surface that matches your toolchain: PG wire for the broadest compatibility (Python, Go, Rust, BI tools); JDBC for JVM-native data pipelines.
18+ translator passes rewrite Postgres idioms into Oracle equivalents on the fly: LIMIT n → FETCH FIRST n ROWS ONLY, ILIKE → UPPER() LIKE UPPER(), the regex operators ~/!~/~*/!~* → REGEXP_LIKE, date_trunc → TRUNC, EXCEPT → MINUS, WITH RECURSIVE preserved, TRUE/FALSE mapped to 1/0 or 'Y'/'N' depending on the column's actual Oracle type. pg_catalog and information_schema are emulated against an embedded DuckDB catalog so client schema-introspection queries never touch Oracle.
ofrag's semantic search operates in two modes. The lightweight mode uses keyword matching, synonym expansion, and business process cross-referencing with zero external dependencies. The enhanced mode adds vector embeddings (Gemini Embedding 2) across 25,000+ table descriptions and 500+ REST API resources, understanding queries like "money received from customers" even when exact terms don't match any table name.
Yes. ofrag is an MCP Server that connects AI agents directly to Oracle Fusion Cloud. The AI agent can search for relevant tables using natural language, validate SQL queries against cached schemas, and execute them in real-time — typically returning results in about 4 seconds.
MCP (Model Context Protocol) is an open standard that lets AI agents use external tools. ofrag implements MCP to provide Claude, Gemini, and other LLMs with tools like semantic_search, execute_oracle_sql, rest_call, lint_sql, and more — giving the AI structured access to Oracle Fusion data while keeping production databases protected.
Three free open-source paths, depending on your toolchain. ofpgproxy makes Oracle Fusion look like Postgres — use any PG client (pgAdmin, DBeaver, psql, Metabase, DuckDB, dbt, etc.). OFJDBC is a JDBC driver for DBeaver and JVM tools. ofrag is the AI-powered surface: it lets agents like Claude and Gemini write, validate, and execute SQL through natural language. All three share the same BI Publisher backend mechanism.
Both ofrag and ofpgproxy use local intelligence — metadata is cached locally and exploration (schema search, SQL validation, semantic parsing, pg_catalog probes) never touches the production database. Only final, validated queries reach the live Oracle Fusion system. ofpgproxy is read-only by design — DML paths simply don't exist in the codebase.
Yes. OFJDBC is a standard JDBC driver and works in any JVM-based platform including Java, Kotlin, Scala, and Groovy. It also supports ETL tools like Apache Airflow (JdbcOperator), Talend, Pentaho, and Apache NiFi. For non-JVM languages, ofpgproxy offers PG wire which natively binds to Python (psycopg, asyncpg, SQLAlchemy, pandas), Go (pgx), Rust (tokio-postgres), and every other Postgres client driver.
Yes — three of them, all open source: ofpgproxy (PostgreSQL wire proxy), OFJDBC (JDBC driver), and ofrag (AI/RAG). Together they cover every modern data-tool surface and provide direct SQL query access to Oracle Fusion Cloud, matching commercial tools like SQLConnect or DataFusing without licensing fees. Combined with ofrag, you also get AI-powered querying — something no commercial tool offers.
Yes. ofrag, ofpgproxy, and OFJDBC all connect directly from your machine to Oracle Fusion Cloud. No data passes through third-party servers. ofrag processes all metadata exploration locally; ofpgproxy answers pg_catalog probes from a local DuckDB cache without touching Oracle; OFJDBC source code is fully available on GitHub for review.