GNETiX Docs
Architecture

LLM Providers

Configure your AI model provider

Each GNETiX organization configures its own LLM provider and credentials. No data is shared between tenants, and API keys are encrypted at rest using Fernet encryption.

Supported Providers

Anthropic

Direct access to Claude models via the Anthropic API.

FieldRequiredDescription
API KeyYesYour Anthropic API key (starts with sk-ant-)

Model mapping:

TierModel
fastClaude Haiku 4.5
balancedClaude Sonnet 4.6
powerfulClaude Opus 4.6

Anthropic is the only direct-API provider that supports extended thinking, giving Claude a dedicated reasoning budget for complex tasks.

OpenAI

Direct access to GPT models via the OpenAI API.

FieldRequiredDescription
API KeyYesYour OpenAI API key (starts with sk-)

Model mapping:

TierModel
fastGPT-4o-mini
balancedGPT-4o
powerfulGPT-4o

Amazon Bedrock

Access Claude models through your AWS account. Data stays within your VPC and is subject to your AWS policies.

FieldRequiredDescription
AWS Access Key IDYesIAM access key with Bedrock invoke permissions
AWS Secret Access KeyYesCorresponding secret key
AWS RegionYesRegion where Bedrock is enabled (e.g., us-east-1)

Model mapping:

TierModel
fastClaude Haiku 3.5
balancedClaude Sonnet 4
powerfulClaude Sonnet 4

Bedrock supports extended thinking since it runs Claude models under the hood.

Azure OpenAI

Access GPT models through your Azure subscription. Suitable for organizations with enterprise compliance requirements.

FieldRequiredDescription
API KeyYesAzure OpenAI resource key
Endpoint URLYesYour resource endpoint (e.g., https://my-resource.openai.azure.com)

Model mapping:

TierModel
fastGPT-4o-mini
balancedGPT-4o
powerfulGPT-4o

Configuration

To configure an LLM provider for your organization:

  1. Navigate to Org Settings in the sidebar.
  2. Open the LLM tab.
  3. Select your provider from the dropdown.
  4. Enter the required credentials for your chosen provider.
  5. Save the configuration.

The API key (or AWS credentials) are encrypted immediately on save. The UI only ever displays a masked hint (e.g., ***...a1b2) -- the plaintext key is never retrievable after initial entry.

Only one provider can be active per organization at a time. Switching providers is instantaneous and takes effect on the next message.

Model Tiers

Rather than selecting specific models, GNETiX uses a tier abstraction. The Director requests a tier (fast, balanced, or powerful) and the system maps it to the appropriate model for your configured provider.

  • fast -- optimized for low latency and simple tasks (classification, short answers, routing)
  • balanced -- general-purpose tier used by default for most conversations
  • powerful -- maximum capability for complex reasoning, multi-step analysis, and nuanced responses

Tier-to-model mappings are defined in models.yaml and can be updated without redeploying the backend.

Extended Thinking

Extended thinking gives Claude a dedicated internal reasoning budget before producing its visible response. This improves performance on complex multi-step problems, tool selection, and nuanced analysis.

ProviderExtended Thinking Support
AnthropicYes
Amazon BedrockYes
OpenAINo
Azure OpenAINo

When enabled, the Director allocates a 5,000-token thinking budget on each LLM call. This is configured per-organization in the Org Settings > LLM tab via the extended thinking toggle.

Spend Tracking and Budget Caps

Every LLM call is logged with token counts and estimated cost. GNETiX tracks spend per-organization, enabling:

  • Per-org spend dashboards -- view token usage and cost breakdown over time
  • Budget caps -- set a maximum USD spend per organization; when the cap is reached, LLM calls are blocked until an admin raises or removes the limit

Budget caps are configured in the Org Settings > LLM tab. Setting the budget to empty removes the cap entirely.