Model Management CLI

Manage AI models directly in the database via CLI commands. No UI interaction required.

Commands

Add a new model

bash

bun run model:add --json '{
  "modelId": "openai:gpt-6",
  "provider": "openai",
  "name": "GPT-6",
  "contextWindow": 200000,
  "inputCost": 2.5,
  "outputCost": 10,
  "capabilities": ["vision", "function-calling"],
  "status": "active",
  "userFriendlyDescription": "Next generation GPT with advanced capabilities",
  "bestFor": "Complex reasoning, research, advanced multimodal tasks",
  "knowledgeCutoff": "January 2026"
}'

List models

bash

# All models
bun run model:list

# Filter by provider
bun run model:list --provider openai
bun run model:list --provider anthropic

# Filter by status
bun run model:list --status active
bun run model:list --status deprecated
bun run model:list --status beta

Get a single model

bash

bun run model:get --id "openai:gpt-5"

Update model

bash

# Update pricing
bun run model:update --id "openai:gpt-5" --json '{"inputCost": 2.0, "outputCost": 8.0}'

# Update with reason
bun run model:update --id "openai:gpt-5" --json '{"inputCost": 2.0}' --reason "Price reduction announced"

# Update multiple fields
bun run model:update --id "openai:gpt-5" --json '{
  "inputCost": 2.0,
  "outputCost": 8.0,
  "contextWindow": 250000,
  "userFriendlyDescription": "Updated description"
}'

Deprecate model

bash

# Simple deprecation
bun run model:deprecate --id "openai:gpt-4"

# With reason
bun run model:deprecate --id "openai:gpt-4" --reason "Replaced by GPT-5"

Required Fields (for add)

Field	Type	Description
`modelId`	string	Unique ID: "provider:model-name"
`provider`	string	openai, anthropic, google, xai, perplexity, groq, cerebras, minimax, deepseek, kimi, zai, meta, mistral, alibaba, zhipu
`name`	string	Display name
`contextWindow`	number	Max tokens
`inputCost`	number	Cost per 1M input tokens (USD)
`outputCost`	number	Cost per 1M output tokens (USD)
`capabilities`	array	vision, function-calling, thinking, extended-thinking, image-generation
`status`	string	active, deprecated, beta

Optional Fields

Field	Type	Description
`description`	string	Internal description
`userFriendlyDescription`	string	User-facing description
`bestFor`	string	Use case recommendations
`knowledgeCutoff`	string	Knowledge cutoff date
`cachedInputCost`	number	Cost per 1M cached input tokens
`reasoningCost`	number	Cost per 1M reasoning tokens
`reasoningConfig`	string	JSON string for reasoning configuration
`gateway`	string	"vercel" or "openrouter"
`hostOrder`	array	Fallback hosts (e.g., ["cerebras", "groq"])
`actualModelId`	string	Override for API calls
`speedTier`	string	"ultra-fast", "fast", "medium", "slow"
`isPro`	boolean	Requires tier access
`isInternalOnly`	boolean	Hidden from model picker
`isExperimental`	boolean	Preview/beta flag
`isLocal`	boolean	Ollama models
`benchmarks`	string	JSON string for benchmark scores

Full Example: Adding a New OpenAI Model

bash

bun run model:add --json '{
  "modelId": "openai:gpt-6",
  "provider": "openai",
  "name": "GPT-6",
  "description": "Next generation flagship with advanced reasoning and multimodal capabilities",
  "contextWindow": 400000,
  "inputCost": 3.0,
  "outputCost": 12.0,
  "cachedInputCost": 0.3,
  "capabilities": ["thinking", "vision", "function-calling"],
  "status": "active",
  "reasoningConfig": "{\"type\":\"openai-reasoning-effort\",\"effortMapping\":{\"low\":\"low\",\"medium\":\"medium\",\"high\":\"high\"},\"summaryLevel\":\"detailed\",\"useResponsesAPI\":true}",
  "knowledgeCutoff": "January 2026",
  "userFriendlyDescription": "Most powerful GPT-6. Handles the most complex tasks with advanced reasoning, vision, and deep thinking.",
  "bestFor": "Complex reasoning, research, advanced multimodal tasks",
  "speedTier": "medium"
}'

Full Example: Adding a Free OpenRouter Model

bash

bun run model:add --json '{
  "modelId": "openrouter:new-free-model",
  "provider": "deepseek",
  "name": "New Free Model",
  "description": "Free model via OpenRouter",
  "contextWindow": 128000,
  "inputCost": 0,
  "outputCost": 0,
  "capabilities": ["thinking"],
  "status": "active",
  "gateway": "openrouter",
  "actualModelId": "deepseek/new-model:free",
  "reasoningConfig": "{\"type\":\"deepseek-tag-extraction\",\"tagName\":\"think\",\"applyMiddleware\":true}",
  "userFriendlyDescription": "Free reasoning model",
  "bestFor": "Complex reasoning, experimentation"
}'

Reasoning Config Templates

OpenAI (Reasoning Effort API)

json

{
  "type": "openai-reasoning-effort",
  "effortMapping": { "low": "low", "medium": "medium", "high": "high" },
  "summaryLevel": "detailed",
  "useResponsesAPI": true
}

Anthropic (Extended Thinking)

json

{
  "type": "anthropic-extended-thinking",
  "budgetMapping": { "low": 5000, "medium": 15000, "high": 30000 },
  "betaHeader": "interleaved-thinking-2025-05-14"
}

Google (Thinking Budget)

json

{
  "type": "google-thinking-budget",
  "budgetMapping": { "low": 4096, "medium": 12288, "high": 24576 }
}

DeepSeek (Tag Extraction)

json

{
  "type": "deepseek-tag-extraction",
  "tagName": "think",
  "applyMiddleware": true
}

Key Files

•packages/backend/convex/models/cli.ts - Internal mutations for CLI
•scripts/model-cli.ts - CLI wrapper script
•packages/backend/convex/models/mutations.ts - Admin UI mutations (auth required)
•packages/backend/convex/schema/models.ts - Database schema

History Tracking

All changes are automatically tracked in modelHistory table:

•Version number increments on each change
•Change type: created, updated, deprecated, reactivated
•Field-level changes with old/new values
•Timestamp and reason

Common Workflows

New model announced

•Check existing: bun run model:list --provider openai
•Add model: bun run model:add --json '{...}'
•Verify: bun run model:get --id "openai:new-model"

Model pricing changed

•Get current: bun run model:get --id "openai:gpt-5"
•Update pricing: bun run model:update --id "openai:gpt-5" --json '{"inputCost": 2.0}' --reason "Price reduction"

Model deprecated by provider

•Deprecate: bun run model:deprecate --id "openai:gpt-4" --reason "Replaced by GPT-5"
•Verify: bun run model:list --status deprecated