rayvise-project

rayvise-cli Project Context

Architecture Overview

CLI Tool: Cobra-based CLI for generating AI-optimized prompts via OpenRouter API with automatic project context awareness

Key Packages:

• cmd/ - Cobra commands (root, generate, interactive)
• internal/config/ - Viper config + model registry
• internal/llm/ - OpenRouter client with streaming
• internal/prompts/ - Prompt templates + store
• internal/projectcontext/ - Project context detection and loading
• internal/clipboard/ - Cross-platform clipboard ops
• pkg/types/ - Shared types

Core Concepts

Output Lengths: Control response size via max_tokens + system prompt directives

• short: 300 tokens, concise prompts
• medium: 800 tokens, balanced detail
• long: 1500 tokens, comprehensive

Model Resolution: Custom models → DefaultModels → treat as OpenRouter ID

Config Hierarchy: CLI flags → env vars (RAYVISE_*) → ~/.rayvise/config.yaml → defaults

Streaming: SSE parsing for interactive mode, non-streaming for generate command

Project Context Awareness: Automatically discovers and loads context from .cursor/rules/, CLAUDE.md, or AGENTS.md files (in priority order). Context is injected into the system prompt via {{.Context}} template variable, allowing generated prompts to be aware of project conventions and architecture.

Code Patterns

Error Handling:

// Network errors: retry once
if err != nil || resp.StatusCode >= 500 {
    time.Sleep(1 * time.Second)
    resp, err = client.Do(req)
}

// User-facing errors: clear, actionable
return fmt.Errorf("API key not found. Set RAYVISE_API_KEY or add to config.yaml")

Config Access:

cfg := config.GetConfig()
apiKey := cfg.GetAPIKey()  // Falls back to env var
model := cfg.GetDefaultModel()  // Falls back to "cerebras-llama-8b"

Prompt Rendering:

store, _ := prompts.NewStore()
systemPrompt, _ := store.Render("metaprompt", types.OutputLengthMedium)
// Template {{.LengthDirective}} auto-replaced
// Template {{.Context}} auto-replaced with project context (if found)