API Catalog

Separation of Duties: See Separation of Duties - API documentation is responsible for documenting APIs. It does NOT own agent manifests, execution, or infrastructure configuration.

Complete listing of all APIs across the Bluefly LLM Platform ecosystem.

Overview

Total APIs: 240+ services across 6 categories Architecture: OpenAPI-First, RESTful, gRPC where applicable Documentation: All services maintain OpenAPI 3.1 specifications Authentication: Unified JWT/OAuth 2.0 with GitLab integration

API Categories

1. Agent Services (12 APIs)

Core agent communication, coordination, and intelligence services.

OpenAPI Specs:

• Agent Router API
• Agent Mesh API
• Agent Tracer API
• Agent Protocol API

---

2. Workflow Engines (4 APIs)

Workflow orchestration, compliance, and automation engines.

OpenAPI Specs:

• Workflow Engine API
• Compliance Engine API

Key Features:

• Visual workflow builder (Langflow integration)
• Real-time compliance monitoring
• FedRAMP Moderate baseline support
• NIST AI RMF 1.0 implementation
• Automated audit logging

---

3. Local Models (26 FREE Models)

Zero-cost local inference via Ollama MCP integration.

Featured Models:

• qwen2.5-coder:32b (19GB) - Best for code generation
• gpt-oss:120b (65GB) - General purpose reasoning
• deepseek-coder-v2:16b (9GB) - Fast code completion
• codellama:34b (19GB) - Meta's code model
• gov-rfp-model (4.4GB) - Government RFP analysis
• bfrfp-llm_platform_model (7.4GB) - Platform custom model

API Endpoints:

POST   /invoke           # MCP tool invocation
GET    /tools            # List available tools/models
POST   /api/generate     # Direct inference
GET    /api/tags         # List installed models
GET    /health           # Health check

Cost Savings: 96-99% reduction vs. Claude/OpenAI APIs

• Before: $500-2000/month (API costs)
• After: $20/month (Cursor Pro only)

OpenAPI Spec: /technical-guide/openapi/agent-protocol/local-models-mcp-api.yaml

---

4. Gateway & Routing (2 APIs)

Unified API gateway and intelligent request routing.

Gateway Capabilities:

• Multi-provider routing (OpenAI, Anthropic, Google, Cohere, Ollama)
• Intelligent load balancing
• Response caching
• Rate limiting (1000 req/min per key)
• Cost optimization routing
• Automatic failover

Supported Providers:

• OpenAI (GPT-4, GPT-3.5, DALL-E, Whisper)
• Anthropic (Claude 3, Claude 2)
• Google (Gemini Pro, PaLM)
• Cohere (Command, Embed)
• Ollama (26 local models)
• Custom models (GovRFP, Agent BuildKit)

---

5. Drupal Module APIs (30+ Endpoints)

REST APIs for Drupal custom modules.

Base URL: http://drupal.local.bluefly.io:8080

OpenAPI Specs: /technical-guide/openapi/drupal-modules/

---

6. Build Tools & Infrastructure (8 APIs)

Development tooling and infrastructure management.

---

API Integration Patterns

MCP Routing Pattern

How services call local models through MCP protocol:

graph LR
    A[Service] --> B[Agent Protocol MCP Coordinator]
    B --> C[Ollama MCP Tool Server]
    C --> D[Select Best Model]
    D --> E[Return Result $0 cost]

Example Request:

POST /invoke
{
  "tool": "generate_code",
  "input": {
    "prompt": "Create TypeScript service",
    "model": "qwen2.5-coder:32b"
  }
}

Response:

{
  "result": {
    "output": "...",
    "model": "qwen2.5-coder:32b"
  },
  "cost": 0,
  "duration": 2.5
}

---

Authentication

All APIs use unified authentication via Compliance Engine:

• JWT Tokens: Bearer authentication for service-to-service
• OAuth 2.0: GitLab OAuth for user authentication
• API Keys: For external integrations
• Rate Limiting: Configurable per service

See Authentication Guide for details.

---

OpenAPI Registry

Master Registry: /technical-guide/openapi/master-api-registry.yaml

All OpenAPI specs are:

• OpenAPI 3.1 compliant
• Type-safe with generated TypeScript types
• Validated with Redocly
• Synchronized with live services
• Published to GitLab wiki

---

API Versioning

All APIs follow semantic versioning:

• v1.x.x: Production stable
• v0.x.x: Development/beta
• Breaking changes: Major version bump required

---

Rate Limits

Default rate limits per service:

Rate limits configurable via environment variables.

---

Health Monitoring

All services expose standardized health endpoints:

GET /health

Response:

{
  "status": "healthy",
  "version": "0.4.9",
  "uptime": 86400,
  "dependencies": {
    "database": "healthy",
    "redis": "healthy",
    "mcp": "healthy"
  }
}

---

Next Steps

• Agent Mesh API
• Agent Tracer API
• LLM Gateway API
• MCP Registry API
• Authentication
• Security Architecture