TRAXX

Dify AI: Agentic Procurement Intelligence

Not a chatbot that guesses. An autonomous AI agent with 50+ tools that queries live procurement data, respects your permission model, and takes action — with human oversight where it matters.

How It Actually Works

TRAXX integrates with Dify using a tool-callback architecture. The AI decides what to do — your app executes it securely, scoped to the user's permissions.

User Chat Query + Context SSE Stream TRAXX Session Context Injection Permission Scoping Hallucination Guards Response Sanitization API Call Dify Agent LLM Reasoning Tool Selection Multi-Step Chains RAG Knowledge Base Data Return Tool Callback 50+ Tool Endpoints Procurement · Assets · Contracts Vendors · Finance · Budgets Audit · Reports · Analytics Bearer Token Auth + Audit Log
50+
Tool Endpoints
Procurement, assets, contracts, vendors, finance, budgets, audits, reports — each a secure REST endpoint the AI can call.
SSE
Streaming Responses
Real-time Server-Sent Events. See the AI think, select tools, and build its answer — not a spinning loader followed by a wall of text.
Per-User
Permission Scoping
AI sees only what the user can see. Permission sets control which modules, data, and actions each user's AI agent can access.

Session Context Injection

Every AI query is automatically enriched with the user's identity, permissions, and organizational context. The AI doesn't guess who you are — it knows, and it respects your access boundaries.

User Identity
Name, email, employee ID, department, location — injected into every query so the AI personalizes responses.
Permission Scoping
Allowed modules derived from Permission Sets. A finance user's AI sees budgets and invoices. A warehouse user's AI sees assets and GRN.
Temporal Context
Current date, month, fiscal year — so "this quarter" and "last month" always resolve correctly.
Formatting Rules
Output format instructions (markdown tables, currency format, date format) ensure consistent, readable responses.
Data Integrity Rules
Explicit instructions to never fabricate data, always cite sources, and ask clarifying questions when uncertain.
// What the AI receives with every query:
[Logged-in user]
Name: Priya Sharma
Employee ID: EMP-1042
Location: Mumbai HQ
Allowed modules: procurement, assets, vendors
[Context]
Today: 2026-04-05
Fiscal Year: FY 2026-27
[Rules]
Never fabricate data.
Use tools for all data queries.
Ask clarifying questions if unsure.
> "Show me overdue POs for Q4"

Tool Callback Architecture

The AI doesn't access your database directly. It calls secure tool endpoints — each authenticated, audited, and scoped to the user's permissions. The tool executes the query and returns structured data.

Procurement
PO Status & History
GRN Verification
Invoice Matching
RFQ Comparisons
Rate Card Lookup
PR Approval Status
Asset Management
Asset Search & Details
Depreciation Schedule
Maintenance History
Audit Campaign Status
Transfer Tracking
Disposal Lot Status
Vendors & Contracts
Vendor Performance
Compliance Scores
Contract Terms
AMC Status
Vendor KYC Check
Delivery Tracking
Finance & Reports
Budget vs Actuals
ITC Reconciliation
Spend Analytics
Department Summaries
Export Reports
Chart Generation
// Dify calls TRAXX tool endpoint:
POST /api/dify/tools/po_status
Authorization: Bearer ••••••••
{ "user_id": 1042, "status": "overdue", "quarter": "Q4" }
// Tool executes query scoped to user's permissions
// Returns structured JSON → AI formats for user
// Every call logged to ai_audit_log

Dual Agent Architecture

Two completely isolated AI agents — one for your team, one for your vendors/customers. Different tools, different data access, different API keys. Zero cross-contamination.

Internal Agent

For your procurement, finance, and asset management teams. Full access to internal data scoped by permission sets.

Procurement data & PO management
Asset lifecycle & depreciation
Vendor performance & compliance
Budget tracking & spend analysis
Contract terms & audit findings
Employee-scoped data access
Customer / Vendor Portal Agent

For your vendor portal users. Strictly isolated — sees only their own tickets, orders, AMC status. Cannot access internal data.

Own ticket status & history
AMC coverage & renewal info
Order status & delivery tracking
Create tickets & request callbacks
Satisfaction ratings
Strict email-based identity enforcement

Enterprise Safety Built In

Not just guardrails — a multi-layer safety architecture that prevents hallucination, enforces data boundaries, and maintains complete audit trails.

Hallucination Guards
If the AI responds to a data query without using a tool, the system intercepts and forces a clarifying question. No made-up numbers.
Permission-Scoped Access
Permission Sets control which modules each user's AI can access. A department head sees department data — not company-wide financials.
Full Audit Trail
Every tool call logged — tool name, user ID, arguments, IP address, timestamp. Compliance-ready for SOX and internal audit.
Response Sanitization
Raw SQL, internal reasoning (<think> blocks), and technical artifacts are stripped before the user sees the response.
Bearer Token Auth
Tool callbacks authenticated with a shared secret. Only your Dify instance can call your tool endpoints — no external access.
Identity Enforcement
Customer portal agent enforces identity via email resolution. The AI cannot impersonate another customer or access cross-tenant data.

Rich Chat Experience

Not just text in, text out. The AI chat supports structured forms, interactive charts, downloadable exports, markdown tables — all embedded inline.

Streaming Responses
Real-time SSE streaming — see the AI think and respond token by token. No waiting for a complete answer.
Conversation Persistence
Save, pin, and revisit conversations. Pick up where you left off. Share insights with colleagues.
Inline Data Tables
Structured data returned as formatted markdown tables — sortable, scannable, ready to act on.
Interactive Forms
When the AI needs structured input (dates, selections, amounts), it renders inline form widgets — not plain text prompts.
Chart Generation
AI generates ECharts visualizations inline — bar charts, pie charts, trend lines — from your live procurement data.
Export & Download
AI can generate downloadable reports (CSV, PDF) with expiring secure links — directly from the chat.
AI
TRAXX AI
Online · 50+ tools available
Show me overdue POs for IT department this quarter
You · 2:34 PM
Used tool: po_status
Found 3 overdue POs for IT department in Q4:
| PO# | Vendor | Amount | Days Late |
| PO-4521 | Dell | ₹8.2L | 12 days |
| PO-4533 | HP | ₹3.1L | 7 days |
| PO-4548 | Cisco | ₹15.4L | 3 days |
Total overdue: ₹26.7L. Shall I escalate to the vendor managers?
AI · 2:34 PM

Deployment Options

Recommended
Self-Hosted
Docker deployment with Weaviate (vector DB), Redis, and your choice of LLM. Data never leaves your network. Full control.
Cloud Hosted
TRAXX-managed Dify instance. We handle infrastructure, updates, and scaling. You focus on building workflows.
Model Agnostic
Use Claude, GPT-4o, Llama 3, Mistral, Qwen, or Ollama for local inference. Switch models without rebuilding anything.

See the AI Agent in Action

Book a demo to see tool callbacks, permission-scoped AI, streaming chat, and real procurement data queries — live.