Inference Data Model

Inference manages three collections: providers, models, and requests. All entities use nanoid(8) identifiers and support both file-based and Convex storage backends.

InferenceProvider

A registered AI provider configuration.

interface InferenceProvider {
  id: string;              // nanoid(8)
  name: string;            // Unique name
  type: ProviderType;      // Provider implementation
  baseUrl?: string;        // Custom API base URL
  defaultModel?: string;   // Default model for this provider
  envVar?: string;         // Environment variable for API key
  created: string;         // ISO 8601
  updated: string;         // ISO 8601
}

type ProviderType = "anthropic" | "openai" | "google";

Default Environment Variables

Type	Env Variable
anthropic	ANTHROPIC_API_KEY
openai	OPENAI_API_KEY
google	GEMINI_API_KEY

InferenceModel

A synced AI model with capability metadata.

interface InferenceModel {
  id: string;                       // nanoid(8)
  providerId: string;               // Parent provider ID
  modelId: string;                  // Provider's model identifier
  name: string;                     // Display name
  contextWindow?: number;           // Max context length (tokens)
  maxOutput?: number;               // Max output length (tokens)
  supportsStreaming: boolean;       // SSE streaming support
  supportsEmbeddings: boolean;      // Text embedding support
  supportsImageGeneration: boolean; // Image generation support
  supportsTranscription: boolean;   // Audio transcription support
  created: string;                  // ISO 8601
}

InferenceRequest

A logged inference operation for usage tracking.

interface InferenceRequest {
  id: string;              // nanoid(8)
  providerId: string;      // Provider that handled the request
  model: string;           // Model used
  type: RequestType;       // Operation type
  inputTokens: number;     // Input token count
  outputTokens: number;    // Output token count
  durationMs: number;      // Request duration in milliseconds
  status: "success" | "error";
  error?: string;          // Error message (on failure)
  timestamp: string;       // ISO 8601
}

type RequestType = "chat" | "embed" | "image" | "transcribe";

InferenceConfig

Service configuration.

interface InferenceConfig {
  version: string;
  defaultProvider?: string;  // Default provider ID
  defaultModel?: string;     // Default model name
}

Storage

File-Based (SHIFT_STORAGE=file)

.inference/
  config.json               # InferenceConfig
  providers/<id>.json        # Provider records
  models/<id>.json           # Synced model records
  requests/<id>.json         # Request logs

Convex (SHIFT_STORAGE=convex)

Tables use the inference_ prefix:

Table	Description
inference_providers	Provider configurations
inference_models	Synced model metadata
inference_requests	Request usage logs

Each record includes a sid field (shift ID) that maps to the application-level id.