LLM Provider Management & Routing
The LLM Provider feature enables Tuluat to integrate, manage, and dynamically route completion requests across multiple downstream Large Language Model backends.
1. Supported Providers
| Provider | Type Identifier | Description | Default Model |
|---|---|---|---|
| OpenAI | OPENAI |
Direct OpenAI REST API endpoints (gpt-4o, gpt-4o-mini). |
gpt-4o |
| DeepSeek | DEEPSEEK |
DeepSeek API endpoints (deepseek-chat, deepseek-r1). |
deepseek-chat |
| Ollama | OLLAMA |
Local, zero-cost self-hosted LLM runtime (llama3.2, mistral). |
llama3.2 |
| WireMock | WIREMOCK |
Deterministic stub server for offline E2E acceptance testing. | wiremock-gpt-4o |
| Anthropic | ANTHROPIC |
Anthropic Claude API endpoints (claude-3-5-sonnet). |
claude-3-5-sonnet |
2. Model Gateway & Fallback Chains (ModelGateway)
The ModelGateway handles route resolution, fallback execution, and expenditure tracking:
graph TD
AgentCall[Agent Invocation Request] --> PrimaryCheck{Primary Provider Available?}
PrimaryCheck -- Yes --> ExecPrimary[Call Primary Provider]
PrimaryCheck -- No / Fails --> CheckFallback{Has Configured Fallbacks?}
CheckFallback -- Yes --> ExecFallback[Invoke Fallback Provider in Order]
CheckFallback -- No --> ThrowError[Throw ModelGatewayException]
ExecPrimary --> CostTrack[Record Input & Output Token Spend]
ExecFallback --> CostTrack
CostTrack --> ReturnResponse[Return GatewayCallResult]
2.1 Fallback Specification Example
spec:
providerType: OPENAI
baseUrl: https://api.openai.com/v1
defaultModel: gpt-4o
fallbacks:
- providerName: deepseek-provider
namespace: tuluat-system
model: deepseek-chat
- providerName: ollama-provider
namespace: tuluat-system
model: llama3.2
3. Token Pricing & Expenditure Tracking
Each LlmProvider declares input and output token costs:
- costPer1kInputTokens: Price in USD per 1,000 prompt tokens.
- costPer1kOutputTokens: Price in USD per 1,000 completion tokens.
The operator calculates per-session expenditure and exposes metrics via Prometheus and the Control Portal dashboard.