30k-Star API Gateway vs 265-Star Enterprise Middleware: LinkMind or NewAPI?

When choosing AI middleware, many people come across both NewAPI and LinkMind.

At first glance they seem to do the same thing — unifying APIs from multiple model providers so business systems don’t have to write adapters for each vendor individually. But digging into the code and documentation reveals they are solving problems at completely different levels.

NewAPI (QuantumNous/new-api) is a unified checkout counter for model APIs, focused on one thing: converting APIs from OpenAI, Claude, Gemini and others into a unified interface, with billing, user management, and routing built in. Written in Go, 30.2k Stars.

LinkMind (landingbj/LinkMind) is a unified access layer for enterprise AI capabilities, doing a lot more: chat, OCR, speech-to-text, text-to-speech, image and video generation, RAG, Text-to-SQL, knowledge graphs… all packed into a single Java middleware. 265 Stars, just getting started.

This is not a “which one is better” question, but a “what are you missing” question.

Positioning: Checkout Counter vs Capability Supermarket

NewAPI’s positioning is narrow but precise. Its core feature list is straightforward:

• API format conversion: OpenAI ↔ Claude Messages ↔ Google Gemini
• Multi-channel management: One instance connects multiple model channels, routing by weight
• Billing & user management: Token statistics, usage-based pricing, online top-up (EPay/Stripe)
• Smart routing: Automatic retry on failure, user-level rate limiting

No RAG, no OCR, no speech capabilities. It is just a gateway — but built deeply.

LinkMind’s positioning is entirely different. It aims to become that “unified layer” when enterprises adopt AI:

• Unified chat entry point (overlaps with NewAPI)
• RAG + vector search: Supports Chroma, Elasticsearch, Milvus, MySQL, Pinecone, SQLite
• OCR capabilities: Document-to-structured-data conversion
• ASR/TTS: Speech transcription and text-to-speech
• Image & video capabilities: Image generation, image/video understanding
• Text-to-SQL: Natural language to database queries
• MCP support: Model Context Protocol
• Agent runtime integration: Built-in configuration sync for OpenClaw, Hermes Agent, DeerFlow
• Safety guardrails: Sensitive word filtering, stop-word control
• Cascading networking: Multiple LinkMind nodes form a network with physical and logical isolation

This is a capability supermarket, not a single-purpose checkout counter.

Architecture: Java Full Stack vs Go Lightweight Gateway

NewAPI Tech Stack:

• Go language, high-performance gateway
• SQLite or MySQL storage
• One-click Docker deployment, port 3000
• Compatible with the original One API database
• 5,703 commits, 654 version tags

LinkMind Tech Stack:

• Java, runs on JDK 8+
• Modular architecture: lagi-core (core), lagi-extension (extensions), lagi-web (web layer)
• Supports JAR direct run, Docker, one-click install script
• Unified configuration in lagi.yml
• 897 commits, 17 version tags

Go gives NewAPI a natural advantage in concurrency and memory footprint, making it suitable for high-traffic API gateways. Java gives LinkMind easier integration with existing enterprise Java ecosystems and middleware.

Feature Comparison

Capability	NewAPI	LinkMind
Unified chat entry	✅	✅
OpenAI-compatible API	✅	✅
Claude/Gemini compatibility	✅	✅
Multi-model routing	✅ Weighted random	✅ Airank routing orchestration
API format conversion	✅ Three-way conversion	✅
Failure retry	✅	✅
Token statistics	✅	✅
Billing system	✅ EPay/Stripe	✅ Built-in billing
User management / permissions	✅	✅
Multi-language frontend	✅ 5 languages	❌
RAG / Vector search	❌	✅ 6 vector databases
OCR	❌	✅
ASR/TTS	❌	✅
Text-to-SQL	❌	✅
MCP support	❌	✅
Agent framework integration	❌	✅ 3 frameworks
Sensitive word filtering	❌	✅
Medusa cache acceleration	❌	✅
Graph enhancement	❌	✅
Cascading networking	❌	✅
Midjourney/Suno	✅	❌

Licensing: Caution Required for Both

NewAPI: AGPLv3

This is the AGPLv3 license, meaning if you modify the code and offer it as a service, you must open-source your modifications. Many enterprise legal departments are sensitive to AGPL. The project team also offers a commercial licensing option ([email protected]).

LinkMind: Custom LinkMind License

This is not a standard open-source license. The README states “this project follows the LICENSE,” but uses a custom license. Enterprises should carefully review the terms before use to confirm compliance with their company’s open-source policy requirements.

Both projects’ licenses require special attention from enterprise users.

Use Cases

Reasons to choose NewAPI:

1. You manage API keys for multiple models and need a unified entry point for business systems
2. You need to allocate different quotas and model permissions to different users
3. You need OpenAI ↔ Claude ↔ Gemini format conversion
4. You have high traffic and need a high-performance Go gateway
5. You need a mature, stable project (30k+ Stars, 5,700 commits)

Reasons to choose LinkMind:

1. Your enterprise needs simultaneous access to multiple AI capabilities (chat + OCR + speech + video)
2. You have RAG requirements and need vector search and knowledge-base Q&A
3. You are already using OpenClaw or Hermes Agent and need a unified middleware layer
4. You need enterprise compliance features like sensitive word filtering and content safety
5. You want to evaluate whether a “one middleware covers all AI capabilities” approach is viable

Using both together:

This is not actually an either-or choice. NewAPI handles API gateway and billing, while LinkMind handles multi-modal capability access and Agent runtime. The two can sit at different layers in your architecture.

Maturity Comparison

This is the point that needs to be stated most directly.

NewAPI forks from One API (MIT license), inheriting the original project’s user base and feature accumulation. 30.2k Stars, 6.5k Forks, 5,703 commits, 654 version tags — this is a widely used and actively maintained project. The community is active and documentation is thorough.

LinkMind has only 265 Stars, 15 Forks, 897 commits, and 17 version tags. The code already covers a significant number of capability modules, but the project is still in its early stages overall. The community is small and documentation is relatively lightweight.

If your project needs to go into production tomorrow, NewAPI is the safer choice. If you are doing technology planning and want to evaluate future directions, LinkMind’s approach is worth tracking.

One-Sentence Summary

• NewAPI = Unified checkout counter for model APIs — narrow but deep
• LinkMind = Unified supermarket for enterprise AI capabilities — broad but shallow

Need a gateway? Pick NewAPI. Need middleware that connects chat, OCR, speech, RAG, and Agents all together? Look at LinkMind.