Oracle Fusion RAG Engine (ofrag)

Start chatting with your ERP.

This is a specialized Retrieval-Augmented Generation (RAG) engine designed for LLMs (like Claude). It empowers your AI agent to retrieve schema context locally, validate intent via semantic parsing, and fetch live business data directly from Oracle Fusion Cloud.

---

The Magic Moment

User: "Show me the top 5 unpaid invoices for vendor 'Acme Corp' generated last month."

Claude + ofrag:

1. Understands "unpaid" means PAYMENT_STATUS_FLAG = 'N'.
2. Lints the SQL against local metadata to ensure the table AP_INVOICES_ALL exists.
3. Executes the query directly against your live Oracle Fusion instance.
4. Returns a perfectly formatted markdown table.

Total Time: 4 seconds.

---

Why ofrag?

🚀 Velocity for Experts, Access for Everyone

We shift the paradigm of data access.

• For the Analyst: Data is no longer locked in technical silos. If you can type a question, you can get an answer.
• For the Consultant: This is your productivity multiplier. Skip the 15-minute cycle of logging into BI Publisher, creating data models, and debugging XML. Execute, iterate, and verify complex queries in seconds. It shifts your focus from "how do I extract this?" to "what does this mean?"

🧠 Agentic SQL Synergy (RAG Augmentation)

This isn't just a "dumb pipe" for SQL—it is a cognitive force multiplier for your AI.

• The Workflow: When you ask the LLM to fix or improve a SQL query, it utilizes the lint_sql tool.
• Deep Parsing: The tool parses the provided SQL into an Abstract Syntax Tree (AST) locally, checking it against the Oracle schema cache.
• Multiplied Intelligence: Instead of guessing, the LLM receives precise structural feedback (e.g., "The column VENDOR_NAME does not exist in table AP_INVOICES_ALL"). This augments the LLM's context, allowing it to perform highly accurate SQL repairs and optimizations that would otherwise be hallucinations and tokens waste.

🌐 Universal REST — Self-Discovering API Agent

The rest_call tool isn’t limited to a handful of hardcoded endpoints — it turns your AI into a universal Oracle Fusion REST client.

• Schema-Free Operations: Oracle Fusion REST endpoints are largely self-describing. Claude can call /describe on any resource to understand its structure on the fly, then perform the actual operation — creates, updates, queries — without needing hardcoded schemas.
• Full CRUD: GET, POST, PUT, PATCH, DELETE — the agent handles any HTTP method with managed SSO authentication, so it can read invoices, create suppliers, update PO lines, or call custom endpoints.
• SQL Fallback: When the agent needs deeper data insight (e.g., checking table relationships, resolving IDs, or aggregating across modules), it seamlessly falls back to direct SQL via the existing SQL tools. REST for operations, SQL for analytics — best of both worlds.
• Bulk Export: For large datasets, rest_call_to_file automatically paginates through Oracle Fusion’s {items, hasMore, next} pattern and merges everything into a single local file.

⚡ Local Intelligence (RAG Retrieval)

We respect your environment by running sophisticated analysis locally.

• Advanced Introspection Tools: It’s not just a cache; it’s a search engine. ofrag ships with a suite of local tools—Fuzzy Search (search_identifiers), Semantic Discovery (semantic_search), and Module Context Analyzers (module_summary)—that traverse your schema instantly.
• Vector Search + REST API Catalog (optional): This is a completely optional add-on — semantic_search works fully without it using keyword matching, synonym expansion, and business process cross-referencing. To unlock deeper results, drop the pre-built embeddings.db and rest_catalog.db from the release next to your metadata.db and add a free Gemini API key. This enables Gemini Embedding 2 to understand the meaning of your query across 25K+ SQL table and view descriptions and 500+ REST API resources. Ask for "money received from customers" and it finds both AR_RECEIVABLE_APPLICATIONS_ALL (SQL) and relevant REST endpoints — even though those words don’t appear in any name. See Semantic Search Setup for details.
• Zero-Latency Reasoning: When the AI explores your schema to understand how AP_INVOICES relates to PO_HEADERS, it uses these local tools to "think" about your data structure.
• Less Pressure: Your Oracle Fusion database receives only the final, polished queries—never the heavy exploratory workload.

---

🔄 Multi-Environment Management

Switch between Oracle Fusion environments (dev, SIT, UAT, prod) on the fly — no config edits, no restarts.

• Interactive Setup: Add environments with add_environment — specify name, host URL, and auth type (SSO or basic) directly in the agent conversation.
• Instant Switching: switch_environment(name="prod") — all subsequent queries hit the new environment immediately.
• Cross-Environment Comparison: Use the environment parameter on any query to compare data across environments without switching (e.g., compare tax rates in dev vs prod in parallel).
• Mixed Auth: Each environment can use its own auth method — SSO for prod, basic auth for automation.

See Multi-Environment Management for full details.

Documentation

• Installation & Setup - Detailed guide on how to install, configure, and get the binary running.
• Detailed MCP Server Docs - Complete reference for configuration, authentication, variables and tools.
• Multi-Environment Management - Configure and switch between dev, UAT, prod environments interactively.
• Semantic Search Setup - How to enable vector search and REST API catalog for full-powered semantic discovery.

---