Documentation Index
Fetch the complete documentation index at: https://mintlify.com/paramveer-cyber/Deployaar/llms.txt
Use this file to discover all available pages before exploring further.
Deployaar supports four AI providers — Anthropic, OpenAI, Google Generative AI, and DeepSeek — all accessed through the Vercel AI SDK. Every step of the pipeline that calls an AI model (clarification questions, PRD generation, task breakdown, the coding agent loop) resolves the active language model through a single abstraction in packages/services/ai/index.ts. The active provider and model are determined per-user: Deployaar checks the user’s saved preferences first, then falls back to the platform-wide DEFAULT_AI_PROVIDER and DEFAULT_AI_MODEL environment variables. Access to specific providers and models is gated by your billing plan.
Supported Providers
| Provider | SDK Package | Default Model | Notes |
|---|
| Anthropic | @ai-sdk/anthropic | claude-sonnet-4-6 | Available on Pro Max plans only (unless user supplies their own API key) |
| OpenAI | @ai-sdk/openai | gpt-4o | Available on Pro and Pro Max plans |
| Google Generative AI | @ai-sdk/google | gemini-2.0-flash | Available on all plans, including Free |
| DeepSeek | @ai-sdk/deepseek | deepseek-chat | Available on Pro Max plans only |
packages/services/ai/index.ts via resolveLanguageModelForUser(userId). This function:
- Loads the user’s saved
preferredProvider and preferredModel from user-settings
- Resolves the user’s billing plan via their primary organisation
- Asserts that the requested provider and model are permitted for that plan (unless the user has supplied their own API key, in which case plan restrictions are bypassed)
- Decrypts the user’s stored API key if present, or falls back to the platform key for that provider
- Returns a fully constructed
LanguageModel instance ready to pass to generateText or streamText
Per-Plan AI Access
Access to AI providers and models is enforced in packages/services/billing/limits.ts via the planLimitsMap. When a user does not supply their own API key, assertProviderAllowedForPlan() is called and will throw a FORBIDDEN error if the requested provider or model is outside the plan’s allowlist.
| Plan | Allowed Providers | Allowed Models |
|---|
| Free | Google | gemini-2.0-flash |
| Pro | Google, OpenAI | gemini-1.5-pro, gpt-4o |
| Pro Max | Anthropic, OpenAI, Google, DeepSeek | All models (no model restriction) |
allowedModels: [] (an empty array), which the enforcement logic interprets as “all models permitted” for any allowed provider.
Attempting to use a provider or model that is not included in your plan will throw a FORBIDDEN error with the message: “Your [Plan Name] plan does not include access to the ‘[provider]’ provider. Upgrade to unlock more models.” This check runs before any AI call is made, so no tokens are consumed. To bypass plan restrictions, supply your own API key in your account settings — your personal key is never subject to plan-based model limits.
Configuration
Set the following environment variables in apps/api/.env to configure AI provider access for the platform:
# Select the default provider for all users who haven't set a preference
# Accepted values: anthropic | openai | google | deepseek
DEFAULT_AI_PROVIDER=google
# Default model string for the selected provider
DEFAULT_AI_MODEL=gemini-2.0-flash
# API keys — only the key(s) for your active provider(s) are required
ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...
GOOGLE_GENERATIVE_AI_API_KEY=AIza...
DEEPSEEK_API_KEY=sk-...
DEFAULT_AI_MODEL is not set, the service falls back to the hardcoded default for the selected provider:
| Provider | Hardcoded default |
|---|
anthropic | claude-sonnet-4-6 |
openai | gpt-4o |
google | gemini-2.0-flash |
deepseek | deepseek-chat |
PR Review Model
The AI pull request review step (packages/services/pull-request-review) resolves its language model through the same resolveLanguageModelForUser abstraction used by every other pipeline step. This means the review model respects the platform’s DEFAULT_AI_PROVIDER and DEFAULT_AI_MODEL environment variables and is subject to the same per-plan provider restrictions.
// packages/services/pull-request-review/index.ts (simplified)
const reviewModel = await resolveLanguageModelForUser(requestedByUserId);
const reviewResult = await generateText({
model: reviewModel,
output: Output.object({ schema: aiReviewOutputSchema }),
system: reviewSystemPrompt,
prompt: reviewPrompt,
});
DEFAULT_AI_PROVIDER and DEFAULT_AI_MODEL in apps/api/.env. The review service also defines an unused resolveReviewModel() helper that constructs an Anthropic claude-sonnet-4-6 instance directly — this function is present in the source but is not called by the live review flow.
For a full comparison of what each billing plan includes — feature request limits, project limits, and team seats — see the Billing Plans reference.