AI Gateway

const response = await fetch("/projects/{id}/ai/chat/stream", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({ model: "gpt-4o", messages })
});

const reader = response.body.getReader();
const decoder = new TextDecoder();

while (true) {
  const { done, value } = await reader.read();
  if (done) break;
  const chunk = JSON.parse(decoder.decode(value));
  console.log(chunk.chunk); // append to UI
}

curl -X POST https://api.zmesh.in/api/ai/chat \
  -H "x-api-key: zb_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini",
    "messages": [
      { "role": "user", "content": "Hello!" }
    ]
  }'

{
  "messages": [
    { "role": "user", "content": "Explain REST vs GraphQL" }
  ],
  "models": ["gpt-4o", "claude-sonnet-4", "gemini-2.0-flash"]
}

// Response — parallel results from all models
{
  "results": [
    {
      "model": "gpt-4o",
      "provider": "openai",
      "content": "REST uses resource-based URLs...",
      "latency_ms": 1100,
      "tokens": { "prompt": 12, "completion": 200 }
    },
    {
      "model": "claude-sonnet-4",
      "provider": "anthropic",
      "content": "The key differences between...",
      "latency_ms": 980,
      "tokens": { "prompt": 12, "completion": 185 }
    },
    {
      "model": "gemini-2.0-flash",
      "provider": "google",
      "content": "REST and GraphQL are both...",
      "latency_ms": 650,
      "tokens": { "prompt": 12, "completion": 170 }
    }
  ]
}

{
  "input": "Hello world",
  "model": "text-embedding-3-small"
}

// Response
{
  "embeddings": [[0.0023, -0.0094, 0.0156, ...]],
  "dimensions": 1536,
  "model": "text-embedding-3-small"
}

// 1. Enable pgvector extension
POST /projects/{project_id}/ai/vectors/setup

// 2. Create a collection
POST /projects/{project_id}/ai/vectors/collections
{
  "name": "documents",
  "dimensions": 1536
}

// 3. List collections
GET /projects/{project_id}/ai/vectors/collections

// 4. Upsert documents (auto-embeds text)
POST /projects/{project_id}/ai/vectors/collections/documents/upsert
{
  "documents": [
    {
      "content": "zMesh is an AI-first BaaS platform",
      "metadata": { "type": "docs", "page": "intro" }
    },
    {
      "content": "Edge Functions run in sandboxed subprocesses",
      "metadata": { "type": "docs", "page": "functions" }
    }
  ],
  "model": "text-embedding-3-small"
}

// 5. Semantic search
POST /projects/{project_id}/ai/vectors/collections/documents/search
{
  "query": "How do serverless functions work?",
  "limit": 5,
  "filter": { "type": "docs" }
}

// Response
{
  "results": [
    {
      "id": "...",
      "content": "Edge Functions run in sandboxed subprocesses",
      "similarity": 0.92,
      "metadata": { "type": "docs", "page": "functions" }
    }
  ]
}

{
  "message": "How do I set up RLS for my posts table?"
}

// Zy auto-gathers:
// - Your database schema (tables, columns, types)
// - Auth configuration (providers, settings)
// - Storage buckets and policies
// Then responds with project-specific guidance

// Response
{
  "response": "Based on your schema, your posts table has...",
  "context_used": ["schema", "auth"],
  "cached": false
}

{
  "prompt": "Show me all users who signed up this week",
  "execute": true
}

// Response
{
  "sql": "SELECT * FROM users WHERE created_at >= NOW() - INTERVAL '7 days'",
  "explanation": "Fetches users created in the last 7 days",
  "results": [
    { "id": "...", "name": "Rahul", "created_at": "2026-03-20T..." }
  ]
}

// Usage stats (last 30 days)
GET /projects/{project_id}/ai/usage?days=30

// Response
{
  "total_requests": 1250,
  "total_tokens": { "prompt": 450000, "completion": 320000 },
  "by_provider": {
    "openai": { "requests": 800, "tokens": 500000 },
    "anthropic": { "requests": 350, "tokens": 220000 },
    "google": { "requests": 100, "tokens": 50000 }
  },
  "recent_logs": [...]
}