Overview
The LLM Configuration feature gives you a structured, centralized way to set up and manage the Large Language Models (LLMs) used within the AI Design Studio. It separates credential management from model configuration, so secrets are entered once and reused across multiple Models – eliminating repetitive setup and reducing the risk of misconfiguration.
Configuration is available under AI Design Studio > Advanced > LLM Configuration module, and is organized into two tabs:
- Vendor – Manage credentials and service platforms.
- Model – Define model parameters and link models to vendors.
LLM Management Modes
Every tenant operates in one of two LLM management modes, which determines who owns and manages the LLM credentials.
| PRAMATA MANAGED (Default) | CUSTOMER MANAGED (On Request) | |
| Credentials | Pre-configured using Pramata's credentials — no setup required on your end. | Your organization provides and manages its own credentials (API keys, access keys, deployment URLs, etc.). |
| Default vendors | One Vendor is automatically created. | No default Vendors. You must create Vendors manually. |
| Vendor editability | Default Vendor is read-only and cannot be modified or deleted. | Full control — create, edit, and delete your own vendors. |
| LLM costs | Covered under Pramata's account. | Your organization bears full responsibility for LLM usage costs. |
| Pramata visibility | Pramata manages and monitors LLM configuration. | Pramata has no visibility into or control over credentials you enter. |
| Custom Service Platform | Not available | Available on request — allows use of your own deployment URL or model gateway. |
Switching modes: The LLM management mode is set to Pramata during tenant onboarding. If you need to switch from Pramata-managed to Customer-managed, please reach out to the Pramata Support team.
Warning: Switching modes will remove all existing Vendor configurations and unlink any Models currently mapped to those Vendors. Plan any mode changes carefully to avoid disruption to active LLM usage.
Vendor Configuration
Vendors hold the credentials for a given service platform. Once a Vendor is created, it can be linked to one or more Models – so credentials only need to be entered once, regardless of how many Models share them.
Service Platforms
When creating a Vendor, you select the service platform that will be used to access your LLMs:
| Platform | Description |
| AWS Bedrock | Access LLMs via Amazon’s managed Bedrock service using your AWS credentials. |
| Anthropic | Direct access to Anthropic-hosted Models via the official Anthropic API. The deployment URL defaults to api.anthropic.com. |
| Custom | For organizations hosting their own model gateway or using a proprietary deployment URL. Only available in Customer-managed mode, and requires domain pre-approval – see below. |
Custom Service Platform & Domain Approval
If your organization routes LLM traffic through its own infrastructure (for example, a self-hosted Model gateway or an enterprise proxy), you can use the Custom service platform to point to your deployment URL.
To use this option, your organization must first provide the list of approved domains to the Pramata Support team. These domains are registered against your tenant, and only URLs belonging to those domains will be accepted when creating a Vendor. This ensures that LLM traffic is directed only to endpoints your organization has explicitly authorized.
How to enable Custom: Contact the Pramata Support team with the list of domains you want to allow (Example: abccompany.com). Once registered, you will be able to create Vendors using deployment URLs under those domains.
Creating a Vendor
To create a new Vendor, go to AI Design Studio > Advanced > LLM Configuration > Vendor and click Add New Vendor.
| FIELD | ANTHROPIC | AWS BEDROCK | CUSTOM |
| Name | A label for this vendor. Max 100 characters. | A label for this vendor. Max 100 characters. | A label for this vendor. Max 100 characters. |
| Managed By | Reflects your tenant's management mode. Read-only. | Reflects your tenant's management mode. Read-only. | Reflects your tenant's management mode. Read-only. |
| Service Platform | Select Anthropic | Select AWS Bedrock. | Select Custom |
| Provider | Anthropic. Read-only. | Anthropic. Read-only. | Anthropic. Read-only. |
| Authorization Header | Not applicable | Not applicable | The header key used for authentication (e.g. api-key). |
| Secret key / Password | Your Anthropic API key. | Your AWS secret access key. | The secret or password for your custom endpoint. |
| Deployment URL | Defaults to https://api.anthropic.com/. Read-only. | Not applicable | Your organization's deployment URL. Must match a pre-approved domain registered with Pramata Support. |
| AWS Access Key ID | Not applicable | Your AWS access key ID for Bedrock authentication. | Not applicable |
| AWS Region | Not applicable | The AWS region where your Bedrock service is hosted (e.g. us-east-1). | Not applicable |
| API version | v1. Read-only. | bedrock-2023-05-31. Read-only. | v1. Editable. |
| Anthropic version | 2023-06-01 Read-only. | Not applicable | 2023-06-01. Editable. |
Default Vendor
This setting applies only when Managed By is set to Customer.
In Pramata-managed mode, a default Vendor is pre-configured automatically and cannot be modified. When operating in Customer-managed mode, the first Vendor you create is automatically designated as the default Vendor. The default Vendor:
- Cannot be deleted (it serves as the system’s fallback).
- Can be changed at any time by marking another Vendor as the default.
- Once a new default is set, the previous default can be deleted.
Model Configuration
The Model Configuration tab is where you define your LLM Models and link them to a Vendor. A Model represents a specific LLM (such as Claude Sonnet) together with its runtime parameters.
Supported Models Reference
The table below lists all available Model configurations by Name, Vendor, and Model ID. Refer to this table when filling in the Model / Inference Model field while creating or editing a Model (see Creating a Model below).
| Name | AWS Bedrock Model ID | Anthropic Model ID |
| Balanced | global.anthropic.claude-haiku-4-5-20251001-v1:0 | claude-haiku-4-5-20251001 |
| Deep Digitization | global.anthropic.claude-sonnet-4-6 | claude-sonnet-4-6 |
| Deep Reasoning | global.anthropic.claude-sonnet-4-6 | claude-sonnet-4-6 |
| Digitization | global.anthropic.claude-haiku-4-5-20251001-v1:0 | claude-haiku-4-5-20251001 |
| Expert Reasoning | global.anthropic.claude-opus-4-6-v1 | claude-opus-4-6 |
| Light Reasoning | global.anthropic.claude-haiku-4-5-20251001-v1:0 | claude-haiku-4-5-20251001 |
| Negotiator Deep Reasoning | global.anthropic.claude-sonnet-4-6 | claude-sonnet-4-6 |
| Negotiator Light Reasoning | global.anthropic.claude-haiku-4-5-20251001-v1:0 | claude-haiku-4-5-20251001 |
| Negotiator Reasoning | global.anthropic.claude-sonnet-4-6 | claude-sonnet-4-6 |
| Reasoning | global.anthropic.claude-sonnet-4-6 | claude-sonnet-4-6 |
| Speedy | global.anthropic.claude-haiku-4-5-20251001-v1:0 | claude-haiku-4-5-20251001 |
Creating a Model
To create a new Model, go to AI Design Studio > Advanced > LLM Configuration > Model and click Add New Model.
Note: If you are using AWS Bedrock under a customer-managed configuration, the available model names may differ from Pramata's standard model reference. Refer to your AWS Bedrock console for the exact model IDs available in your account.
| FIELD | DESCRIPTION |
| Name | A display name for the model. |
| Vendor | Select the Vendor whose credentials will be used to access this Model. Only Vendors compatible with your tenant’s configuration are shown. |
| Model / Inference Model | The specific Model variant. Enter the Model ID corresponding to the selected Vendor. Refer to the Supported Models Reference table above for the correct Model ID values. |
| Max output tokens | The maximum number of output tokens the AI can generate in a single response. |
| Maximum Context Window Size | The token limit for the context window. |
| Context window overflow warning threshold | 80 (The percentage of the context window at which a warning is triggered.) |
Note: Some fields are pre-populated and read-only — including Effective Context Window Overflow Warning Threshold and Temperature. Options such as Prompt Caching, Reasoning, and Allow Batching are available only if supported by the selected model and permitted by your service platform.
Model Validation
When a Model is first created or linked to a Vendor, its status is set to Unverified, and a ⚠ warning icon is displayed alongside the Model name. This indicates that the Model’s connection to the linked Vendor has not yet been confirmed. An unverified Model cannot be trusted for LLM execution until its access is confirmed.
Verification can be triggered in any of the following ways:
- Re-save the Model without changes — Open the Model in edit mode and save it without making any changes. The system will re-attempt validation and confirm access to the linked Vendor.
- Run an Agent linked to the Model — Executing any Agent that is already configured to use this Model will automatically trigger verification. If the Agent runs successfully, the Model is marked as verified and the warning icon is removed.
Vendor Deletion and Model Unlinking
If a Vendor is deleted, all Models linked to it are automatically unlinked. Unlinked models display a warning icon and cannot be used for LLM execution until they are re-mapped to a valid Vendor.
To restore an unlinked Model, edit it and select an available Vendor. The warning icon will be cleared once the new Vendor’s credentials are validated.