AI Ignition Kit for IBM i

Solution report card

Runs on IBM i?	Partially (connectivity components run on IBM i; AI workloads run on Linux on IBM Power)
On-prem	✅
IBM Cloud	✅ (Power Virtual Server)
AI capabilities	Generative AI, RAG, document digitization, summarization, similarity search, entity extraction, translation
Commercial support	Varies by component
Free to try?	✅
Requirements	RHEL 9.6+ (ppc64le) with Podman, or Red Hat OpenShift on IBM Power; IBM i Open Source

What is the AI Ignition Kit?

Your business data already lives on IBM i. The fastest path to useful AI is to bring AI to the data. The AI Ignition Kit is a set of components that work together to do exactly that: deploy AI services alongside your IBM i partitions and move data in both directions between IBM i and those services.

The kit has two parts:

• AI Launchpad (workload manager) — deploys, runs, and manages AI workloads, such as AI models and pre-built AI services, on IBM Power.
• Connectivity toolkit (moves data both ways):
  • Db2 SDK (IBM i → AI) — SQL functions and procedures that let Db2 and RPG call AI models, event streams, and messaging services.
  • Mapepire (AI → IBM i) — a secure WebSocket database server that lets AI apps in any container or cloud query Db2 for i.

AI Launchpad

AI Launchpad deploys, runs, and manages AI workloads on IBM Power. It provides:

• Pre-built services — production-ready AI services, from RAG chat to document digitization.
• One control plane — a catalog UI, CLI, and APIs to deploy, monitor, and remove apps.
• Any inference — plug in IBM Spyre, Red Hat AI Inference Server, or any OpenAI-compatible endpoint.

Architecture

Layer	Components
Interfaces	Catalog UI (one-click deploy), ai-services CLI (bootstrap, catalog, apps), Catalog API (automation and bring-your-own-service), MCP sidecar (tools for AI agents)
AI services (containers)	Digital assistant, Digitize, Summarize, Similarity search, Entity extraction, Translation
Data layer	Vector database (OpenSearch), S3 object storage, SSH/SFTP file systems
Inference endpoints	IBM Spyre accelerator for Power, Red Hat AI Inference Server, OpenAI/watsonx.ai-compatible APIs (on-prem or cloud)

AI Launchpad runs on RHEL 9.6+ LPARs with Podman, or on Red Hat OpenShift on IBM Power. Remote workers connect over outbound mTLS.

Pre-built AI services

Service	What it does
Digital assistant	Conversational RAG over your documents
Digitize	Ingest documents, with duplicate detection
Summarize	Condense long text into key points
Similarity search	Semantic search across your content
Entity extraction	Extract structured fields: sync, async, or batch
Translation	Sync and batch, English and German first

Up and running

Three CLI steps take an IBM Power LPAR (RHEL 9.6+ on ppc64le, with Podman or OpenShift) to a running AI app:

1. Bootstrap the LPAR. Validates the OS and IBM Spyre configuration.

   ai-services bootstrap --runtime podman

2. Configure the catalog. Stands up the catalog service and UI.

   ai-services catalog configure --runtime podman

3. Deploy applications. Use one click in the catalog UI, or the CLI.

   ai-services application --help

Once the catalog is configured, an AI service can be deployed from the catalog with one click, on premises or on Power Virtual Server.

Upgrading from v0.3.x?

Back up the OpenSearch vector database and Digitize data first, then follow the v0.3.x → v0.4.0 migration guide in the AI Launchpad documentation.

What’s new in v0.4.0

AI Launchpad v0.4.0 became generally available in October 2026. New in this release:

• Remote workers — deploy to remote Power LPARs and OpenShift clusters from one catalog host over mTLS.
• Data connectors — sync from S3 object storage and SSH/SFTP file systems.
• Bring your own service — onboard custom models and services as bundles via the API and CLI.
• MCP sidecar — expose chat, digitize, summarize, and search to AI agents.
• Granite 4.1 8B on CPU and IBM Spyre.
• Mid-deployment cancellation.

Connectivity toolkit

Db2 SDK: call AI from SQL

The Db2 SDK is an open-source SDK of SQL functions and procedures that lets RPG and Db2 workloads integrate with external services:

• AI inference — IBM watsonx, Ollama, OpenAI-compatible endpoints, Wallaroo
• Events, messaging, and system — Kafka, Slack, SMS (Twilio), PASE

Install it on IBM i:

git clone https://github.com/IBM/AI-SDK-Db2-IBMi
cd AI-SDK-Db2-IBMi
gmake     # builds library DBSDK_V1

Before you start

The installing user profile needs QIBM_DB_SECADM function usage. The SDK is experimental and not yet intended for production use.

Point Db2 at your AI endpoint

Because AI Launchpad services and inference servers expose OpenAI-compatible APIs, you can point Db2 at them directly. First, configure the endpoint for your job:

-- Configure the OpenAI-compatible endpoint for this job
CALL dbsdk_v1.openai_compatible_setserverforjob('ai.example.com');
CALL dbsdk_v1.openai_compatible_setportforjob(443);
CALL dbsdk_v1.openai_compatible_setprotocolforjob('https');
CALL dbsdk_v1.openai_compatible_setmodelforjob('granite');

Use the ...forme procedures instead to persist settings for your user profile. Add an API key with ...setapikeyforjob.

Then call the model from a query:

-- Classify customer feedback where it lives
SELECT f.id, f.comment,
       dbsdk_v1.openai_compatible_chat_generate(
         'Positive, Neutral or Negative? ' || f.comment,
         '{"max_tokens": 10, "temperature": 0.1}') AS sentiment
  FROM mylib.feedback f;

chat_generate posts to /chat/completions and sends only the options you set, so it works with strict endpoints.

Patterns you can ship from SQL

• Enrich — classify, tag, or score rows inside a SELECT, right where the data lives.
• Summarize — roll data up with LISTAGG, then ask a model for insights on the result.
• Persist and notify — INSERT generated content into tables; push events to Kafka, Slack, or SMS.

Mapepire: AI apps meet Db2 for i

Mapepire (pronounced “MAH-pup-ee”) is a database access layer for Db2 for i built on secure WebSockets. It offers:

• A consistent client SDK across languages
• No native drivers on the client
• One port, with data always encrypted
• Support for containers, notebooks, and clouds

	JDBC	ODBC	Mapepire
Needs only a single port			✅
Data is always encrypted			✅
Manageable via exit points	✅	✅	✅
Enhanced CCSID support	✅		✅
Runs in watsonx.ai notebooks			✅
Runs in lightweight containers	✅		✅
Multiple client languages			✅

Connect AI in two steps

1. On IBM i, once per system:

   # Install from the IBM i open source RPMs
   yum install mapepire-server service-commander
   
   # Start once; it keeps running
   sc start mapepire
   
   # Port 8076 by default, always encrypted
   # Access rules: /QOpenSys/etc/mapepire/iprules.conf

2. In your AI workload (Python):

   pip install mapepire-python

   from mapepire_python import connect
   
   creds = {"host": "myibmi", "port": 8076,
            "user": "AIUSER", "password": "…"}
   
   with connect(creds) as conn:
       with conn.execute("select * from sample.employee") as cur:
           print(cur.fetchall())

Mapepire connections are governed by the same exit points as JDBC apps, support allow and deny rules by user and IP address, and accept credentials from .ini files, environment variables, or Kerberos. See Accessing Db2 from Python for more examples.

Putting it together

The components combine into one end-to-end flow:

1. Launch — AI Launchpad deploys and manages the workload, such as a RAG assistant.
2. Send — the Db2 SDK sends IBM i data to the workload from SQL or RPG.
3. Reason — the workload answers, classifies, or acts on the request.
4. Look up — Mapepire lets the workload query Db2 for i securely.

Get started

AI Launchpad documentationInstallation, configuration, and the v0.3.x → v0.4.0 migration guide. Demo: ibm.biz/aiservicestechdemo

AI Launchpad v0.4.0 announcementWhat's new in v0.4.0. Code: github.com/IBM/project-ai-services

Db2 SDKUsage and examples for each supported service.

MapepireServer setup and client guides.