JumpStart SAP Intelligence

JumpStart SAP Intelligence is a Cosmos-owned consulting workbench for blueprint-driven SAP, Oracle, and BTP assessment flows. It helps consultants analyze custom code, plan migrations, assess system configuration, and generate deliverables using Cosmos ADK, Storage, skills, and model configuration.

JumpStart is a standalone Next.js application that connects to Cosmos services through the @cosmos/sdk TypeScript SDK. It has its own database, workbench state, connection credentials, and durable worker process; Cosmos provides identity, project membership and access grants, agent execution, DMS, registry resources, cost limits, and observability.

Current Scope

• Runtime metadata, journey execution, project connections, DMS access, and project RBAC are scoped to the authenticated user and the canonical Cosmos project.
• Project members, groups, and app-surface (journey) grants live in native Cosmos project RBAC. Settings → Project Access shows a journeys × principals matrix with direct, inherited, and effective access.
• Workbench execution passes the execution context (including the manifest hash), skill and knowledge-base selections, and the canonical project ID to ADK through the SDK. ADK owns context compaction and history filtering; JumpStart does not send compaction controls.
• Journeys can show the read-only knowledge bases, skills, and knowledge graphs of their pinned agent version, including an interactive Knowledge Map for graphs.
• SAP workbenches support typed package deployment, diagnostics and job correlation, AXPP files, Smart Forms and Adobe Forms build/deploy, and SAP CPI integration package generation.
• Every chat can export Markdown tables to Excel and extract archive or document contents from DMS files; chat attachments also accept audio and video files, which are transcribed through ADK.

Target Users

How JumpStart Leverages Cosmos

JumpStart is a first-class Cosmos consumer application. It does not implement its own LLM orchestration, tool execution, or authentication — it delegates all of that to the Cosmos platform:

SDK Integration

JumpStart uses the @cosmos/sdk TypeScript package. The SDK provides:

• CosmosClient — orchestrates calls to Configurator, ADK, Storage, RAGFlow, and Observer
• decodeCosmosAccessToken() — stateless JWT decode for extracting user claims
• refreshCosmosToken() — OAuth2 token refresh via User Service
• buildResourceInventory() — resolves activated skills and KB collections into a prompt-ready inventory
• createDocumentTools() — registers file read/write tools for in-conversation document management

import { CosmosClient, decodeCosmosAccessToken } from "@cosmos/sdk";

const client = new CosmosClient({
  orgId: COSMOS_ORG_ID,
  projectId, // canonical Cosmos project UUID, sent as X-Project-Id
  userId: decodeCosmosAccessToken(accessToken).user_id,
  services: {
    user: process.env.COSMOS_USER_SERVICE_URL,
    chat: process.env.COSMOS_CHAT_SERVICE_URL,
    configurator: process.env.COSMOS_CONFIGURATOR_SERVICE_URL,
    tools: process.env.COSMOS_TOOLS_SERVICE_URL,
    observer: process.env.COSMOS_OBSERVER_SERVICE_URL,
    storage: process.env.COSMOS_STORAGE_SERVICE_URL,
    adk: process.env.COSMOS_ADK_SERVICE_URL,
    ragflow: process.env.COSMOS_RAGFLOW_SERVICE_URL,
  },
  gatewayUrl: process.env.COSMOS_GATEWAY_URL, // optional fallback
  authManager: {
    tokens: { accessToken, refreshToken: "", expiresAt },
  },
});

Key Capabilities

1. Blueprint-Driven Workbenches

JumpStart's core UI pattern is the workbench blueprint — a JSON configuration that declaratively defines a multi-panel canvas, the agent prompt, tool registrations, and iterative fix loops. Blueprints let EY teams create new assessment workflows without writing code.

Each blueprint specifies:

• Canvas panels — code editors (Monaco), findings tables, structured documents, forms, diff views, batch progress trackers
• Prompt engineering hints — response format templates for structured LLM output
• Tool exclusions — disable tools irrelevant to the current workflow
• Skill/KB activation — pre-attach relevant skills and knowledge collections
• Iterative fix configuration — automated rounds of AI-assisted code repair with approval gates

2. SAP Code Quality Analysis

The SAP workbench connects to live SAP systems via the ADT (ABAP Development Tools) API:

• Source browsing — search and read ABAP objects by name or package
• ATC quality checks — run SAP's ABAP Test Cockpit from within JumpStart
• Syntax validation — real-time syntax checking before activation
• AI-driven code fixing — LLM analyzes ATC findings and proposes fixes, displayed in a diff view for consultant approval
• Iterative fix loops — up to 3 rounds of automated fix → validate → re-fix, stopping when the code is clean or no further improvement is detected
• Object activation — push validated fixes back to the SAP system

3. Oracle Code Quality Analysis

The Oracle workbench connects to Oracle databases via the oracledb thin-mode driver:

• Object browsing — search PL/SQL objects (packages, procedures, functions, triggers)
• Source viewing — read object source code
• Compile checking — compile objects and surface errors/warnings
• Table DDL inspection — retrieve DDL via DBMS_METADATA
• Dependency analysis — trace object dependency trees
• Explain plans — generate query execution plans for performance analysis

4. BTP Neo-to-Cloud Foundry Migration

The BTP workbench helps assess SAP BTP Neo applications for migration to Cloud Foundry:

• Application inventory — list deployed Neo applications and descriptors
• Dependency analysis — map Java dependencies and SAP-specific libraries
• Destination configuration — catalog configured destinations for migration planning
• Source inspection — browse Java source files and configuration for migration impact assessment

5. Discovery Workbenches

Discovery workbenches provide high-level assessments using in-memory skill definitions:

• SAP Custom Code Assessment — analyze S/4HANA compatibility, identify performance anti-patterns, flag security risks, and compute complexity metrics
• Oracle Migration Assessment — evaluate schema complexity, PL/SQL portability, data volume, deprecated feature usage, and performance-critical objects

6. Project & Engagement Management

JumpStart includes its own data model (Prisma + PostgreSQL) for tracking consulting engagements:

• Projects — represent client engagements with industry classification, status tracking, and project roles (Project Manager, SME, Consultant). Each project maps to a canonical Cosmos project, and its members and groups are managed in native Cosmos project RBAC.
• Project Access matrix — shows each journey against users, groups, and the organization with direct, inherited, and effective access. Users who can manage project members can restrict a journey for a user or group; a restriction that cannot be enforced (for example access through an org-admin permission, a group, or a managed source) is flagged as an error. Members are paginated with search.
• Connections — encrypted credentials for SAP, Oracle, and BTP systems with health-check monitoring
• Journeys — configurable navigation paths that organize workbenches into logical assessment flows
• Insights — AI-generated findings stored for review, categorization, and export
• Engagement Templates — pre-configured project templates that bundle journeys, connector types, skills, and KB collections for rapid client onboarding
• Feedback Inbox — captures Get Started/access-request feedback and exports it to Excel for project onboarding follow-up

7. Knowledge Map and Agent Resources

The chat composer offers Agent knowledge, Agent skills, and Knowledge Map buttons for journeys with a pinned agent version. They show, read-only, the knowledge bases, skills, and knowledge graphs that agent version uses. The Knowledge Map renders a graph with community and entity views. Settings → Graphs lists the graphs in the project catalog.

A graph is shown only when the journey agent's manifest includes graph.read for it, the graph is ready, and the project catalog grants it. JumpStart rechecks access on every read.

8. Built-in Tools

JumpStart registers some tools in the worker itself rather than through a connector:

Built form artifacts are deployed with the SAP connector tool sap_deploy_form, which checks locks, backs up the existing object to DMS, replaces it, restores the backup on failure, and generates the form.

Technology Stack

Related Documentation

• Architecture — service interaction diagrams and data flow
• Integration Guide — SDK usage patterns and tool creation
• Deployment — deployment approaches and client onboarding
• Cosmos ADK Service — agent execution engine
• Cosmos Configurator Service — tool and agent management
• TypeScript SDK Quickstart — getting started with @cosmos/sdk