I. Core Positioning: The Magic Peng API can uniformly convert dozens of different formats of large model apis into the standard OpenAI interface format, enabling downstream applications to access them without any threshold and achieving "one API access to all AI services". Ii. Main Functional Modules 1. Interface Aggregation and Protocol Conversion Function Description: Unified interface single-point access to all AI services Compatible with OpenAI Standard Format -4 Multi-protocol support supports OpenAI, Claude Messages format, Rerank model (Cohere/Jina) -6 Realtime API supports OpenAI real-time API (including Azure channel) -6 Model format conversion seamlessly converts API formats from different vendors to the standard OpenAI format -2 2. Intelligent Routing and High Availability Function Description: Multi-channel integration integrates multiple API keys into a unified channel pool -2 Load balancing supports dynamic round-robin distribution by weight -2-4 Automatic failover and automatic retry of other channels when a single point of failure occurs -4 Weighted Random Distribution Supports weighted random invocation by channels -6 3. Billing and asset management function description: Detailed billing supports billing by frequency/volume, prepaid recharge -4 Multiple rate configuration for different channels, groups can set different rates -4 Online recharge supports Easy Payment online recharge function -6 Cache billing: When cache hits, it can be charged proportionally (0-1) Supports OpenAI, Azure, DeepSeek, Claude-6. Use key to query the credit limit. In combination with neko-api-key-tool, the usage credit limit can be queried. -6 4. Permission and Security Control Function Description: Token Management supports the creation and management of API access tokens Call quota, concurrent limit and validity period can be set -2-4 Group model limit token can limit the accessible models by group -6 User rate limiting supports model rate limiting function for users -6 Multi-channel authorization login supports LinuxDO, Telegram Multiple login methods such as OIDC -6 Call and audit complete API call logs and audit trails -4 5. Data Monitoring and Kanban Function Description: Real-time data Kanban usage statistics, cost analysis, and request volume trends -4. Multi-channel monitoring of request times, success rate, and consumption quota for each model -2. Multi-language support for multi-language internationalization -6. Data export support for log and statistical data export 6. Chat Interface Function Description: The built-in chat interface supports routing /chat2link to directly enter the chat interface -6 Online test The model can be directly tested and called on the Web end. 7. Description of advanced features and functions of the model Reasoning Effort control support setting the thinking intensity through the suffix of the model name (o series models: -high/-medium/-low) -6 Claude Thinking Mode Add -thinking suffix enable Claude thinking mode -6 Think to Content support convert the thinking process into final content output -6 8. Compatibility and Extended Function Description: Database Compatibility: Compatible with the original One API database, seamless migration is possible. -6 Multi-tenant architecture: Supports individual, team, and enterprise-level deployments. -4 Multi-runtime: Supports Bun, Node.js, and Deno runtime environments. -10 Containerized deployment: Provides Docker images. Support one-click deployment -6. Supported Service providers: Magic Peng API supports over 30 mainstream AI service providers, including -4 International mainstream category service providers include OpenAI, Anthropic Claude, and Google Gemini. Domestic manufacturers such as DeepSeek, Tongyi Qianwen, and Zhipu offer multi-modal Midjourney services. Suno Local Model Ollama (Locally Run open-source Model) -2. Typical Application Scenarios Scenario Description: Team collaboration, unified management of AI API access, setting usage quotas by project/member -3 Cost Control, Multi-channel aggregation + quota management Accurately track consumption across various channels -3. High availability architecture with multi-provider load balancing and failover to ensure service availability -3. Individual developers uniformly manage local Ollama+ cloud apis and integrate them across multiple channels -2. Commercial distribution provides white-label AI API access for downstream customers Tracking usage through the dashboard -3 5. Official extension tool Magic Peng api-skills (AI editor plugin) available in Claude Code, Cursor, In AI encoding assistants like OpenClaw, you can directly invoke the Magic Peng API function command function: /newapi models list available models /newapi groups view groups and quotas /newapi balance query account balances /newapi tokens List API tokens /newapi create-token to create a new token /newapi switch-group to switch token groups