Generative AI Toolkit for SAP HANA Cloud

About this project

Generative AI Client for SAP HANA Cloud is an extension of the existing HANA ML Python client library, mainly focusing on GenAI and related use cases. It includes many leading-edge GenAI related open source libraries and provides seamless integration with HANA ML, HANA vector engine, and other SAP GenAI Hub SDK, see our Introduction, Notebook and Documentation.

Requirements and Setup

The prerequisites for using the Generative AI Toolkit for SAP HANA Cloud are listed at Prerequisites.

The Generative AI Toolkit for SAP HANA Cloud is available as a Python package. You can install it via pip:

pip install hana-ai

ContextAgent

The toolkit includes a file-based ContextAgent for conversational forecasting and data-preparation workflows with Markdown-backed memory, tool calling, and runtime-selectable skills.

Core ContextAgent skills include:

• data_ingestion_and_dataset_preparation for CSV import and time-ordered train, test, and validation table creation
• timeseries_data_profiling for dataset reports and statistical checks before model selection
• timeseries_forecasting for single-series train, predict, score, and plot workflows
• prediction_result_analysis for predicted-versus-actual comparison and quality analysis
• outlier_detection_and_repair_prep for anomaly inspection before model training
• massive_forecasting for grouped forecasting across many related series
• model_lifecycle_and_artifacts for listing, deleting, and packaging saved models
• hana_dataframe_fallback for SQL transformations (Python fallback requires explicit trusted-host opt-in)

Tools

The toolkit exposes HANAML-oriented tools for data preparation, profiling, forecasting, evaluation, artifact generation, and grouped forecasting workflows.

Core forecasting and analysis tools include:

• import_csv_to_table, split_table_for_forecasting, fetch_data
• ts_dataset_report, ts_check, stationarity_test, trend_test, seasonality_test, white_noise_test
• automatic_timeseries_fit_and_save, automatic_timeseries_load_model_and_predict, automatic_timeseries_load_model_and_score
• additive_model_forecast_fit_and_save, additive_model_forecast_load_model_and_predict, intermittent_forecast
• ts_outlier_detection, ts_make_future_table, forecast_line_plot, accuracy_measure
• list_models, delete_models, cap_artifacts, hdi_artifacts
• SelectStatement_to_table

Grouped and massive forecasting tools include:

• massive_ts_check, massive_ts_outlier_detection
• massive_automatic_timeseries_fit_and_save, massive_automatic_timeseries_load_model_and_predict, massive_automatic_timeseries_load_model_and_score
• massive_additive_model_forecast_fit_and_save, massive_additive_model_forecast_load_model_and_predict
• ts_make_future_table_for_massive_forecast

See the ContextAgent notebook for an end-to-end example at nutest/testscripts/demo/e2e_scenarios/context_agent.ipynb.

Python execution trust boundary

python_hanaml_exec is disabled by default and is not included in the default HANAMLToolkit tool list, including used_tools='all', name-based selection, and connection refreshes. Directly constructed instances also refuse to execute code unless the host explicitly opts in. Tool-call arguments cannot enable execution.

This tool is not a sandbox. Code runs in the host process with its filesystem, network, and HANA database permissions. The AST checks and reduced builtins are best-effort guardrails only: full pandas/hana-ml objects support file I/O and raw database access, and sql() does not enforce SQL authorization. There is no execution timeout or resource isolation.

Applications that deliberately run fully trusted code from trusted callers must construct PythonHanaMLExecTool(connection_context=cc, allow_dangerous_code=True) from hana_ai.tools.hana_ml_tools.python_exec_tools and register that instance using toolkit.add_custom_tool(tool). Never expose an enabled instance to untrusted agent or MCP callers. Untrusted-code execution requires a separately designed OS-level isolation boundary, scoped credentials, and resource limits; do not rely on this tool's syntax checks for containment.

SQL agent deployment boundary

The SQL agent requires an explicit schema or table allowlist. For example, pass schema="APP_READ" or include_tables=["SALES_VIEW"] to create_hana_sql_agent() and create_hana_sql_toolkit(). The application rejects queries outside that configured scope, but this parser-level check is defense in depth and does not replace HANA authorization.

Run SQL agents with a dedicated HANA technical user or role that has only the required SELECT privileges on approved schemas, tables, or restricted views. Do not grant DDL/DML, unrestricted cross-schema access, or procedure execution unless the application explicitly needs it. Validate effective privileges against the target HANA deployment as part of release testing. Views and synonyms must be reviewed together with the privileges of their underlying objects.

MCP transport and vector-store security defaults

Environment-built HANA connections and runtime connection updates explicitly enable TLS encryption and certificate validation by default, including omitted/null update options. Configure the required trust material. Runtime admin tools remain host-opt-in and require authentication over HTTP; remote MCP also requires HTTPS termination. The HTTP client's insecure-network override requires the actual Boolean allow_insecure_http=True, not a string such as "false". Stdio profiles must point to trusted executables and arguments; shell interpreters and Windows batch/PowerShell scripts are rejected even when listed in a profile. Keep those executables, scripts, their directories, and the inherited environment under trusted host control.

Vector-store configuration is validated before metadata lookup or table creation. Table/schema names containing single quotes are rejected because the supported hana-ml metadata API interpolates them into SQL literals. Existing double-quote, NUL, and semicolon restrictions still apply. Ordinary case-sensitive names and query text containing apostrophes remain supported through safe encoding.

Support, Feedback, Contributing

This project is open to feature requests/suggestions, bug reports etc. via GitHub issues. Contribution and feedback are encouraged and always welcome. For more information about how to contribute, the project structure, as well as additional contribution information, see our Contribution Guidelines.

Security / Disclosure

If you find any bug that may be a security problem, please follow our instructions at in our security policy on how to report it. Please do not create GitHub issues for security-related doubts or problems.

Code of Conduct

We as members, contributors, and leaders pledge to make participation in our community a harassment-free experience for everyone. By participating in this project, you agree to abide by its Code of Conduct at all times.

Licensing

Copyright 2026 SAP SE or an SAP affiliate company and generative-ai-toolkit-for-sap-hana-cloud contributors. Please see our LICENSE for copyright and license information. Detailed information including third-party components and their licensing/copyright information is available via the REUSE tool.