SAGE-MCP

# Model Configuration for SAGE MCP # This file defines all model capabilities, hints, and selection criteria models: # OpenAI Models o3: provider: openai display_name: "OpenAI o3" emoji: "🧠" capabilities: reasoning: excellent speed: slow context_limit: 256000 # 256K tokens confirmed from benchmarks cost: very_high strengths: - deep_reasoning - mathematical_proofs - complex_debugging - algorithm_design modes: preferred: [think, debug, analyze] suitable: [review, plan] complexity: min: high optimal: very_high description: "Deep reasoning model for complex problems requiring careful step-by-step analysis" hint: "Use for: Mathematical proofs, algorithm design, complex debugging, deep analysis" selection_priority: 1 # Higher priority for complex tasks # Model-specific API parameters api_parameters: max_completion_tokens: 32768 # o3 uses this instead of max_tokens temperature: 1.0 # o3 has fixed temperature # Don't send system messages - o3 doesn't support them no_system_messages: true gpt-5: provider: openai display_name: "OpenAI GPT-5" emoji: "🔧" capabilities: reasoning: very_good speed: medium context_limit: 400000 # 272K input + 128K output = 400K total cost: high strengths: - planning - code_generation - tool_use - refactoring modes: preferred: [plan, refactor, test] suitable: [debug, review, analyze] complexity: min: medium optimal: high description: "Advanced model with excellent tool use and code generation capabilities" hint: "Use for: Project planning, code generation, refactoring, test creation" selection_priority: 2 # Model-specific API parameters api_parameters: max_completion_tokens: 32768 # gpt-5 uses max_completion_tokens instead of max_tokens temperature: 1.0 # GPT-5 should always run at fixed temperature # Google Gemini Models gemini-2.5-pro: provider: gemini display_name: "Gemini 2.5 Pro" emoji: "📚" capabilities: reasoning: very_good speed: medium context_limit: 1000000 # 1M tokens (2M coming soon) cost: medium strengths: - long_context_analysis - comprehensive_reviews - deep_thinking - multi_file_analysis modes: preferred: [analyze, review, think] suitable: [debug, plan, refactor] complexity: min: medium optimal: high description: "Powerful model with massive 2M token context window for comprehensive analysis" hint: "Use for: Large codebases, documentation review, multi-file analysis, long conversations" selection_priority: 3 api_parameters: max_tokens: 32768 # Gemini models use max_tokens gemini-2.5-flash: provider: gemini display_name: "Gemini 2.5 Flash" emoji: "⚡" capabilities: reasoning: good speed: fast context_limit: 1000000 # 1M tokens cost: low strengths: - balanced_performance - general_tasks - quick_analysis - debugging modes: preferred: [chat, debug, refactor] suitable: [analyze, review, test] complexity: min: low optimal: medium description: "Fast model with good reasoning and 1M context, excellent price/performance ratio" hint: "Use for: General development tasks, quick analysis, standard debugging" selection_priority: 4 api_parameters: max_tokens: 32768 # Gemini models use max_tokens gemini-1.5-pro: provider: gemini display_name: "Gemini 1.5 Pro" emoji: "📖" capabilities: reasoning: good speed: medium context_limit: 2000000 # Also 2M but older cost: medium strengths: - stable_performance - long_context modes: preferred: [analyze, review] suitable: [chat, debug, plan] complexity: min: low optimal: medium description: "Legacy long-context model, stable and reliable" hint: "Use for: Fallback when newer models unavailable, stable long context needs" selection_priority: 6 api_parameters: max_tokens: 32768 # Gemini models use max_tokens gemini-1.5-flash: provider: gemini display_name: "Gemini 1.5 Flash" emoji: "🏃" capabilities: reasoning: basic speed: very_fast context_limit: 1000000 cost: very_low strengths: - simple_queries - quick_responses - cost_efficiency modes: preferred: [chat] suitable: [debug, refactor] complexity: min: minimal optimal: low description: "Fastest and cheapest model for simple tasks" hint: "Use for: Simple questions, quick lookups, basic chat, cost-sensitive tasks" selection_priority: 10 api_parameters: max_tokens: 32768 # Gemini models use max_tokens # Anthropic Models (via OpenRouter) claude-opus-4.1: provider: anthropic display_name: "Claude Opus 4.1" emoji: "🎯" capabilities: reasoning: excellent speed: slow context_limit: 200000 cost: very_high strengths: - coding - deep_analysis - complex_reasoning modes: preferred: [analyze, debug, think] suitable: [review, plan, refactor] complexity: min: high optimal: very_high description: "Claude's most capable model, excellent at coding (SWE-bench leader)" hint: "Use for: Complex coding tasks, deep analysis, careful reasoning" selection_priority: 2 api_parameters: max_tokens: 32768 # Anthropic models use max_tokens claude-sonnet-4: provider: anthropic display_name: "Claude Sonnet 4" emoji: "💫" capabilities: reasoning: very_good speed: medium context_limit: 200000 cost: medium strengths: - balanced_capability - efficient_reasoning modes: preferred: [analyze, review, debug] suitable: [chat, plan, refactor] complexity: min: medium optimal: high description: "Balanced Claude model with good performance" hint: "Use for: General development, balanced performance needs" selection_priority: 4 api_parameters: max_tokens: 32768 # Anthropic models use max_tokens # DeepSeek Models deepseek-reasoner: provider: deepseek display_name: "DeepSeek Reasoner" emoji: "🧩" capabilities: reasoning: excellent speed: slow context_limit: 64000 cost: low strengths: - deep_reasoning - mathematical_analysis - complex_problem_solving - cost_effective_reasoning modes: preferred: [think, debug, analyze] suitable: [review, plan] complexity: min: high optimal: very_high description: "DeepSeek's reasoning model for complex problems with step-by-step analysis" hint: "Use for: Complex reasoning, mathematical problems, deep analysis at lower cost" selection_priority: 2 api_parameters: max_tokens: 8192 deepseek-chat: provider: deepseek display_name: "DeepSeek Chat" emoji: "💬" capabilities: reasoning: very_good speed: fast context_limit: 64000 cost: very_low strengths: - general_tasks - code_generation - quick_responses - cost_efficiency modes: preferred: [chat, debug, refactor] suitable: [analyze, review, test, plan] complexity: min: low optimal: high description: "Fast and cost-effective general purpose model with strong coding capabilities" hint: "Use for: General chat, code generation, debugging, refactoring - excellent price/performance" selection_priority: 3 api_parameters: max_tokens: 8192 deepseek-coder: provider: deepseek display_name: "DeepSeek Coder" emoji: "👨‍💻" capabilities: reasoning: very_good speed: fast context_limit: 64000 cost: very_low strengths: - code_generation - code_analysis - refactoring - debugging modes: preferred: [refactor, debug, test] suitable: [analyze, review, plan] complexity: min: low optimal: high description: "Specialized coding model optimized for software development tasks" hint: "Use for: Code generation, refactoring, debugging, test creation" selection_priority: 4 api_parameters: max_tokens: 8192 # Model selection rules selection_rules: context_thresholds: small: 10000 # < 10K tokens medium: 100000 # < 100K tokens large: 500000 # < 500K tokens massive: 2000000 # < 2M tokens complexity_indicators: high: keywords: [algorithm, proof, optimize, architecture, design, complex] file_count: 5 token_count: 50000 medium: keywords: [debug, refactor, analyze, review, test] file_count: 2 token_count: 10000 low: keywords: [explain, what, how, simple, basic] file_count: 1 token_count: 2000 # Provider configurations providers: openai: requires_api_key: true env_var: OPENAI_API_KEY supports_streaming: true supports_tools: true gemini: requires_api_key: true env_var: GEMINI_API_KEY supports_streaming: true supports_tools: true anthropic: requires_api_key: true env_var: ANTHROPIC_API_KEY supports_streaming: true supports_tools: false openrouter: requires_api_key: true env_var: OPENROUTER_API_KEY supports_streaming: true supports_tools: false deepseek: requires_api_key: true env_var: DEEPSEEK_API_KEY supports_streaming: true supports_tools: true