add AI support

Initial Release

Release Notes: iris-vector-rag v0.11.4 (OEX 1.0.2)

This release represents a complete rewrite and production hardening of the
framework since the initial OEX listing (1.0.1).

Package renamed

Previously listed on OEX as intersystems-iris-rag (rag-templates on PyPI).
Now published as iris-vector-rag on PyPI with a matching OEX module name.
Importable as iris_vector_rag. The iris_rag/ shim preserves legacy imports.

Six pipeline strategies, unified interface

All pipelines share one factory call and one response shape — answer,
retrieved_documents, contexts, sources, metadata — swappable with a
single create_pipeline(type, ...) call:

• BasicRAG — dense vector retrieval
• BasicRAGReranking — retrieval + cross-encoder reranking
• CRAG — Corrective RAG with relevance-gated web fallback
• HybridGraphRAG — GraphRAG + vector hybrid with entity extraction
• MultiQueryRRF — multi-query generation fused via Reciprocal Rank Fusion
• ColBERT/PyLate (PLAID) — late-interaction ColBERT with in-DB PLAID index

ObjectScript SDK (RAG.SDK.*)

Five ObjectScript classes callable directly from IRIS without touching Python:

• RAG.SDK.Pipeline — run any pipeline strategy by name
• RAG.SDK.Search — BM25, IVFFlat, and vector search paths
• RAG.SDK.Schema — table initialization, schema status, pip install hook
• RAG.SDK.Bridge — overlay support and default table configuration
• RAG.SDK.Evaluate — RAGAS evaluation from ObjectScript

Connection layer overhauled

• All IRIS connections route through get_iris_connection() — one path,
  no duplicates.
• Auto-detects embedded runtime via iris.runtime.get().state — skips TCP
  entirely when running inside IRIS or when IRISINSTALLDIR is set.
• Lazy import iris.dbapi in connection_pool.py — no ImportError at
  import time when iris is absent.
• Unified via iris-embedded-python-wrapper — handles embedded-kernel,
  embedded-local, and native-remote backends transparently.

attach_existing_corpus

Zero-copy bridge: point any pipeline at tables already in IRIS without
re-ingesting data.

Validation layer

Pre-flight checks before any pipeline runs: required tables exist, embeddings
are ≥95% non-NULL, IRIS VECTOR format valid. auto_setup=True creates missing
tables and embeddings on first use.

RAGAS evaluation

Side-by-side pipeline comparison with faithfulness, context precision, and
context recall. Uses real PMC biomedical documents — no synthetic data.

REST API and MCP server

• FastAPI REST API ([api] extra) with Redis-backed sessions.
• MCP server ([mcp] extra) — all pipelines exposed as MCP tools, usable
  from Claude and other MCP clients.

iris-vector-graph integration

GraphRAG, ColBERT/PLAID, BM25, IVFFlat, and shortestPath delegate to
iris-vector-graph — IRIS-native graph and vector operations without leaving
the database.

IPM/ZPM install

zpm install iris-vector-rag now also runs
pip install iris-vector-rag==0.11.4 via RAG.SDK.Schema.Install().

CI/CD

• GitHub Actions CI on Python 3.11 and 3.12.
• Release workflow: tests → build → PyPI (OIDC trusted publishing) →
  GitHub Release with changelog notes attached.

First release on OEX!

Initial Release

try WebApplication again

Initial Release

module version bump

one more bugfix with folder path

bugfix with WebApplication

Added IPM

Initial Release

Initial Release

Initial Release

Moved main functionality out of a class and into the RunScript routine. Some code golf optimizations to bring this down to 491 characters in total, 441 of which are Python code.

Initial Release

Fixed issue with web setup script

Optimized RunScript for speed by replacing the%SYSTEM.WorkMgr multi-process design with a single embedded-Python
call into a GIL-releasing OpenMP C kernel.

Added

• App.PEXSetup.cls to configure %Java Server for remote jgw and register PEX components
• Runtime PEX setup via pex-runtime.script and docker-iris-start.sh on IRIS container start
• Telegram bot credentials via .env (TELEGRAM_BOT_TOKEN, TELEGRAM_BOT_USERNAME)
• .env.example template for local configuration
• Published Docker images on Docker Hub: jackpto/iris-crypto-tracker-iris and jackpto/iris-crypto-tracker-jgw (tags 1.2.0, latest)

Changed

• Java Gateway runtime upgraded from Java 8 to Temurin 11 (Cloudflare/CoinPaprika TLS compatibility)
• IRIS JDBC/utils JARs aligned with IRIS 2026 (intersystems-jdbc-3.10.5, intersystems-utils-4.2.2)
• CryptoProduction uses %gatewayName=%Java Server instead of host/port settings
• CryptoService HTTP client: explicit User-Agent, Accept header, timeouts, and HTTP status handling
• Docker Compose: IRIS hostname iris, depends_on: jgw, jgw loads .env
• Base Java build image switched to eclipse-temurin:8-jdk

Fixed

• PEX / Java Gateway connection to jgw:55555 (NOTOPEN, gateway name mismatch)
• CoinPaprika API calls failing with SSLHandshakeException / blocked Java 8 User-Agent
• Missing OnTearDown() and invalid super.OnMessage() / super.OnProcessInput() calls in PEX Java classes
• Installer namespace/path for /opt/irisapp/
• Empty CSV header not flushed on first create

Security

• Removed hardcoded Telegram bot token from source; secrets stay in .env (gitignored)

Initial Release

update to 1.0.5

• IRIS BI textMeter widgets: added dataProperty bindings in Overview.dashboard — DSW 4.x only renders measures listed in data properties
• IRIS BI UomTable: replaced removed Country.Uom property with [idwaterresultuom].[H1].[Label] row level (fixes @NOPROPERTY)
• KPI pivots (MeanPollution, MeanOutsideInfluence, VenueCount): measure defined via columnLevel instead of deprecated <measure> element
• Docker build: replaced deprecated PyPI package sklearn with scikit-learn in requirements.txt
• Docker build: removed automatic cube build from ZPM Invokes — failed cube activation rolled back module installation and data
• Flask/Dash app: replaced obsolete app.run_server() with app.run() for Dash 4.x compatibility
• Container startup: replaced exit 1 with wait in server_start.sh so background processes keep running
• DeepSee cube: removed duplicate Uom property from Country level in waterPollutionCube (18 833 build errors)
• IRIS BI / DSW: added SetupDSWAddons() to register and compile DSW.Addons.* portlet classes for worldMap and other addons

Initial Release

Add usage information to the README.

• Added HL7 schema loader for ITB.XML to fix schema import on 2026+

v2.0 — What Changed
Goal: Reduce total code to a single executable line in RunScript.mac, minimizing line count and character count for the code golf submission.

The problem with v1.0. The solution split across two files: load_gaia.py (27 lines of Python to parse, flatten, and load the data) + RunScript.mac (1 executable line with a complex SQL pivot). Total: 28 lines, ~1,910 characters.

What changed in v2.0. load_gaia.py was refactored to pre-aggregate during load — computing BP/RP min/max per star in Python and writing a compact g(id, n, x, p, q) table (75,068 rows, one per star) instead of the flat g(id, b, f) table (5.6M rows). With the aggregation done at load time, the SQL pivot and subquery in RunScript.mac disappeared entirely, replaced by a direct single-table scan.

The solution file is now just RunScript.mac — one file, one executable line:

ROUTINE RunScript
s f=##class(%Stream.FileCharacter).%New(),f.Filename="...variable_objects.csv" d f.WriteLine("source_id,...") s r=##class(%SQL.Statement).%ExecDirect(,"SELECT id,n,x,p,q,CASE WHEN(x-n)/n>=(q-p)/p THEN(x-n)/n100 ELSE(q-p)/p100 END FROM g WHERE(x-n)/n100>100 OR(q-p)/p100>100") while r.%Next(){d f.WriteLine(r.%GetData(1)","..._r.%GetData(6))} d f.%Save()

v1.0 → v2.0

• Files: 2 (load_gaia.py + RunScript.mac) → 1 (RunScript.mac)
• Executable lines: 28 → 1
• Characters: ~1,910 → 516
• SQL complexity: pivot + subquery → single-table scan
• Output: 57,099 ✓ → 57,099 ✓

Attempt to resolve python conflict

DEPRECATED

DEPRECATED

DEPRECATED