Teradata AgentStack SDKs
teradata_agentstack is a suite of Python SDKs developed by Teradata for managing
AgentOps, Access Manager, Inference Engine, MCP Management, RayOps, and related
AgentStack services. It offers a comprehensive set of APIs for deploying and
operating models, agents, evaluations, MCP servers, and Ray jobs on the
Teradata platform.
A blueprint() helper lists all available classes and operations for a module.
For community support, please visit the Teradata Community.
For Teradata customer support, please visit Teradata Support.
Copyright 2026, Teradata. All Rights Reserved.
Table of Contents
- Version Compatibility
- Release Notes
- Included SDKs
- Installation
- Authentication
- Quick Start
- Notebook Workflows
- Pydantic Model Support
- Polling Async Resources
- The blueprint() Function
- Notes and Limitations
- License
Version Compatibility
| AgentStack Version | AI Studio Version |
|---|---|
| v20.00.00.00 | AI Studio 1.3 |
| v20.00.00.01 (Hotfix) | AI Studio 1.3 AWS (with Unstructured) |
| v20.00.00.02 | AI Studio 1.3 Update 1 |
| v20.00.00.03 | AI Studio 1.4 |
Release Notes
Version 20.00.00.03
Platform Target: AI Studio 1.4
-
New Features/Functionality
-
Access Manager
- Added a Playground execute endpoint to try out an API key directly against a model.
-
AgentOps
- Added an Evaluations resource to create, list, view, cancel, and view logs for evaluation runs.
-
Inference Engine
- Added a HuggingFace connectivity check endpoint.
- Added an endpoint to list available GPU types.
- Added an endpoint to view deployment events.
- Added an endpoint to list locally stored models.
-
MCP Management
- Added a server config resource to get, create, update, and delete configuration for an MCP server.
-
RayOps
- Job details, status, logs, and delete are now available by
submission_id.
- Job details, status, logs, and delete are now available by
-
-
Parameter Changes
-
Breaking Changes
-
Access Manager
- Health and usage response fields have changed; review before upgrading.
-
AgentOps
repository_urlwas removed from the Git sync request.
-
Inference Engine
- Model and deployment creation requests have updated field requirements.
-
MCP Management
- The server response shape has changed.
-
RayOps
- Update any code referencing the old
job_name/code_fileparameter names above.
- Update any code referencing the old
No changes were made this release to
agento11y,mcp_endpoint_management,modelops, orsqlgen. -
Version 20.00.00.02
Platform Target: AI Studio 1.3 Update 1
-
Key Changes
- Documentation Refinements: General documentation updates and improvements across the package README.
- Public Notebook Workflows: Added direct reference links to public sample notebook workflows.
Version 20.00.00.01
Platform Target: AI Studio 1.3 AWS (with Unstructured)
-
Key Fixes & Deliverables
- RayOps Startup & Import Fix: Resolved a startup/import issue that could stop the RayOps module /
teradata_agentstacklibraries from loading properly during OPF Ray operations. - Access Manager Fallback Models: Fixed Access Manager fallback models so API responses are recognized correctly and operate reliably.
- Contract-Based Test Generation: Updated the test case generator to support contract-based testing and ran dynamic test suites to validate all test runs.
- Documentation & Notebook Deliverables: Published SDK documentation/HTML reference files and created updated sample notebooks following JupyterHub (JH) guidelines.
- RayOps Startup & Import Fix: Resolved a startup/import issue that could stop the RayOps module /
Version 20.00.00.00
Platform Target: AI Studio 1.3
The initial release of the teradata_agentstack Python SDK, providing a unified interface across 7 core Teradata AI Studio platform services.
-
Supported SDK Modules
agentops(teradata_agentstack.agentops): Package, deploy, version, start/stop, and manage always-on AI agents and deployment secrets.rayops(teradata_agentstack.rayops): Provision node groups and manage the lifecycle of Ray clusters for distributed workloads.agento11y(teradata_agentstack.agento11y): Trace, inspect spans, execute online/offline evaluations, manage datasets, record feedback scores, and aggregate metrics.access_manager(teradata_agentstack.access_manager): Manage user and admin API keys, proxy approved LiteLLM management operations, and inspect tenant usage metrics.inference_engine(teradata_agentstack.inference_engine): Manage model catalogs, versions, workspace secrets, inference deployments (vLLM/KServe/NIM), and async download jobs.mcp_management(teradata_agentstack.mcp_management): Deploy, scale, restart, and manage Model Context Protocol (MCP) server instances on Kubernetes.mcp_endpoint_management(teradata_agentstack.mcp_endpoint_management): Configure custom endpoint path visibility mappings for MCP tools, prompts, and resources.
-
Core Capabilities Introduced
- Interactive Discovery (
blueprint()): Introspect available SDK classes and methods dynamically at runtime. - Unified Authentication: Standardized auth mechanisms across all services with 4 supported modes (
BearerAuth,BasicAuth,ClientCredentialsAuth,DeviceCodeAuth) supporting direct passing, environment variables, or YAML config files. - Pydantic Model Support: Strongly typed request models with automatic serialization.
- Async Polling: Built-in
.poll(id=...)utility for long-running operations with automatic terminal state detection. - Centralized Error Handling: Standardized retry/backoff logic and HTTP status mapping across all modules.
- Interactive Discovery (
Included SDKs
| Module | Client Class | Generated Classes | Service |
|---|---|---|---|
teradata_agentstack.agentops |
AgentOpsClient |
Definitions, Deployments, Secrets, Health |
Teradata AgentOps |
teradata_agentstack.rayops |
RayClusterManagementClient |
RayClusterManagement, ... |
Ray Cluster Management |
teradata_agentstack.agento11y |
AgentO11yClient |
Traces, Projects, Evaluations, ... |
Agent Observability |
teradata_agentstack.access_manager |
AccessManagerClient |
Keys, Admin, Proxy |
API Access Manager |
teradata_agentstack.inference_engine |
InferenceEngineClient |
Secrets, Pools, Endpoints, ... |
Inference Engine |
teradata_agentstack.mcp_management |
MCPManagementClient |
Servers, ... |
MCP Management Service |
teradata_agentstack.mcp_endpoint_management |
MCPEndpointManagementClient |
Endpoints, ... |
MCP Endpoint Management |
Installation
Version: 20.00.00.03 | Python: ≥ 3.9 (64-bit) | Platform: Windows, macOS, Linux
pip install teradata-agentstack
Note: A 64-bit Python environment is required. Installation will fail on 32-bit platforms.
Dependencies
| Package | Version |
|---|---|
teradataml |
>=20.00.00.10 |
requests |
≥ 2.33.0 |
oauthlib |
≥ 3.2.2 |
requests-oauthlib |
≥ 2.0.0 |
pydantic |
≥ 2.10.6 |
PyYAML |
≥ 6.0.2 |
pandas |
≥ 0.22 |
Authentication
All SDK clients accept the same authentication objects, imported from the top-level package:
from teradata_agentstack import ClientCredentialsAuth, DeviceCodeAuth, BearerAuth, BasicAuth
ClientCredentialsAuth — OAuth2 Client Credentials
auth = ClientCredentialsAuth(
auth_token_url="https://your-sso/token",
auth_client_id="your_client_id",
auth_client_secret="your_client_secret"
)
DeviceCodeAuth — OAuth2 Device Code Flow
auth = DeviceCodeAuth(
auth_token_url="https://your-sso/token",
auth_device_auth_url="https://your-sso/device_authorization",
auth_client_id="your_client_id"
)
BearerAuth — Static Token
auth = BearerAuth(auth_bearer="your_bearer_token")
BasicAuth — Username / Password
auth = BasicAuth(username="user", password="password")
Authentication Precedence Order
Each configuration value (base_url, auth, ssl_verify) is resolved in the following order. The first source that provides a value wins:
- Constructor arguments — values passed directly to the client (e.g.,
base_url=...,auth=...). - Environment variables — values read from the process environment (e.g.,
BASE_URL,BASE_API_AUTH_MODE). - YAML config file — values loaded from the config file.
If none of the three sources supplies a required value, a
TeradataMlExceptionis raised.
Authentication via Environment Variables
You can configure authentication without passing an auth object by setting environment variables:
| Variable | Description |
|---|---|
BASE_URL |
Base URL of the API endpoint |
BASE_API_AUTH_MODE |
client_credentials, device_code, bearer, or basic |
BASE_SSL_VERIFY |
true or false |
BASE_API_AUTH_CLIENT_ID |
OAuth2 client ID |
BASE_API_AUTH_CLIENT_SECRET |
OAuth2 client secret |
BASE_API_AUTH_TOKEN_URL |
OAuth2 token endpoint URL |
BASE_API_AUTH_DEVICE_AUTH_URL |
OAuth2 device authorization URL |
BASE_API_AUTH_BEARER_TOKEN |
Static bearer token |
Authentication via YAML Config File
# config.yaml
base_url: https://your-server
ssl_verify: true
auth_mode: client_credentials
auth_client_id: your_client_id
auth_client_secret: your_client_secret
auth_token_url: https://your-sso/token
Pass the path using the config_file argument on any client.
Quick Start
AgentOps SDK
from teradata_agentstack import DeviceCodeAuth
from teradata_agentstack.agentops import AgentOpsClient
from teradata_agentstack.agentops import Definitions, Deployments, Secrets, Health
from teradata_agentstack.agentops.models import (
AgentDefinitionCreateRequest,
DeploymentCreateRequest,
)
auth = DeviceCodeAuth(
auth_token_url="https://your-sso/token",
auth_device_auth_url="https://your-sso/device_authorization",
auth_client_id="your_client_id"
)
client = AgentOpsClient(
base_url="https://your-agentops-server",
auth=auth,
workspace_id="your-workspace-id",
namespace="your-k8s-namespace"
)
# Create API instances
agent_definitions = Definitions(client=client)
deployments = Deployments(client=client)
secrets = Secrets(client=client)
# Create an agent definition — pass Pydantic model directly to body=
created = agent_definitions.create(
body=AgentDefinitionCreateRequest(
name="my-agent",
framework="LangGraph",
artifacts={"storage_type": "zip"},
),
file="/path/to/agent.zip",
)
# Poll until status reaches terminal state (published/failed)
agent_definitions.poll(id=created.id)
# Create a deployment
deployment = deployments.create(body=DeploymentCreateRequest(...))
# Poll until deployment is active/failed/stopped
deployments.poll(id=deployment.agent_id)
Note: The old class names (
AgentDefinitions,AgentDeployments,AgentSecrets) still work as backward-compatible aliases.
Ray Cluster Management SDK
from teradata_agentstack import ClientCredentialsAuth
from teradata_agentstack.rayops import RayClusterManagementClient
auth = ClientCredentialsAuth(
auth_token_url="https://your-sso/token",
auth_client_id="your_client_id",
auth_client_secret="your_client_secret"
)
client = RayClusterManagementClient(
base_url="https://your-ray-server",
auth_data=auth
)
Agent Observability SDK
from teradata_agentstack import BearerAuth
from teradata_agentstack.agento11y import AgentO11yClient
auth = BearerAuth(auth_bearer="your_bearer_token")
client = AgentO11yClient(
base_url="https://your-agento11y-server",
auth=auth,
project_id="your-project-id"
)
Access Manager SDK
from teradata_agentstack import BearerAuth
from teradata_agentstack.access_manager import AccessManagerClient
auth = BearerAuth(auth_bearer="your_bearer_token")
client = AccessManagerClient(
base_url="https://your-server:9876",
auth=auth
)
Inference Engine SDK
from teradata_agentstack import ClientCredentialsAuth
from teradata_agentstack.inference_engine import InferenceEngineClient
auth = ClientCredentialsAuth(
auth_token_url="https://your-sso/token",
auth_client_id="your_client_id",
auth_client_secret="your_client_secret"
)
client = InferenceEngineClient(
base_url="https://your-inference-server",
auth=auth,
workspace="your-workspace"
)
MCP Management SDK
from teradata_agentstack import ClientCredentialsAuth
from teradata_agentstack.mcp_management import MCPManagementClient
auth = ClientCredentialsAuth(
auth_token_url="https://your-sso/token",
auth_client_id="your_client_id",
auth_client_secret="your_client_secret"
)
client = MCPManagementClient(
base_url="https://your-mcp-server",
auth=auth
)
MCP Endpoint Management SDK
from teradata_agentstack import BasicAuth
from teradata_agentstack.mcp_endpoint_management import MCPEndpointManagementClient
auth = BasicAuth(username="user", password="password")
client = MCPEndpointManagementClient(
base_url="https://your-mcp-server:8001",
auth=auth
)
Notebook Workflows
End-to-end Jupyter notebook examples demonstrating common workflows are available in the public repository:
Pydantic Model Support
All SDK modules support passing Pydantic models directly to request body parameters. The SDK automatically serializes models before sending:
from teradata_agentstack.agentops.models import AgentDefinitionCreateRequest
# Pass Pydantic model directly
result = agent_definitions.create(
body=AgentDefinitionCreateRequest(name="my-agent", framework="LangGraph", ...),
file="/path/to/agent.zip",
)
Polling Async Resources
Some SDK classes support a .poll() method for resources that are processed asynchronously:
| Module | Class | Terminal States | Success States |
|---|---|---|---|
agentops |
Definitions |
published, failed |
published |
agentops |
Deployments |
active, failed, stopped |
active |
# Poll until agent definition is published (or failed)
result = agent_definitions.poll(id=agent_def_id, interval=5, max_attempts=20)
# Poll until deployment is active (or failed/stopped)
result = deployments.poll(id=deployment_id, interval=10, max_attempts=30)
Parameters:
id(required): Resource ID to pollinterval(optional): Seconds between polls (default varies by resource)max_attempts(optional): Maximum attempts before timeout
The blueprint() Function
Every SDK module exposes a blueprint() function that lists all available classes and operations for that module:
from teradata_agentstack.agentops import blueprint
blueprint()
# ----------------------------------------------------------------
# Available classes for AgentOps SDK:
# * teradata_agentstack.agentops.Definitions
# * teradata_agentstack.agentops.Deployments
# ...
# ----------------------------------------------------------------
Notes and Limitations
- Only keyword arguments are accepted in API methods; positional arguments are not supported.
- Pydantic models can be passed directly to
bodyparameters — the SDK handles serialization automatically. - TLS certificate verification is enabled by default. Disable only in trusted development environments (
ssl_verify=False).
License
Copyright 2026, Teradata. All Rights Reserved. Teradata Confidential and Trade Secret.
Metadata
Release files for teradata-agentstack 20.0.0.3
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| teradata_agentstack-20.0.0.3-py3-none-any.whl | Python 3 | none | any | Details for teradata_agentstack-20.0.0.3-py3-none-any.whl |
Release files / teradata_agentstack-20.0.0.3-py3-none-any.whl
| Download URL | teradata_agentstack-20.0.0.3-py3-none-any.whl |
|---|---|
| Size | 210.3 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
c548728374f26d95710db637dcacbdd545b57e381eb24c9ec129a7ccc1249096
|
|
BLAKE2b-256 checksum How to use checksums |
054b459ea341f3539d5e1b342081a60553af5c32b7eb7b72ac6cf8f4c11b5e58
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.12
|
Security for teradata-agentstack 20.0.0.3
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.