Getting Started with Aeion AI

Navigate: AI → Providers → Add Provider

The AI module needs provider credentials. Each provider has its own setup:

OpenAI:

json { "providerId": "openai", "name": "OpenAI", "status": "active", "credentials": { "apiKey": "{{OPENAI_API_KEY}}" }, "defaults": { "textModel": "gpt-4o", "imageModel": "dall-e-3", "embeddingModel": "text-embedding-3-small" } }

Anthropic:

json { "providerId": "anthropic", "name": "Anthropic", "status": "active", "credentials": { "apiKey": "{{ANTHROPIC_API_KEY}}" }, "defaults": { "textModel": "claude-sonnet-4-20250514" } }

Google (Gemini on Vertex AI):

json { "providerId": "google", "name": "Google AI", "status": "active", "credentials": { "projectId": "{{GCP_PROJECT_ID}}", "credentials": "{{GCP_SERVICE_ACCOUNT_JSON}}" }, "defaults": { "textModel": "gemini-2.0-flash" } }

All provider credentials are encrypted at rest using a key scoped to the AI layer — never stored in plaintext. Rotation is safe, revocation immediate.

Using the unified AI API:

```typescript // Text completion const result = await ctx.ai.complete({ model: "gpt-4o", messages: [ { role: "system", content: "You are a helpful assistant." }, { role: "user", content: "Summarize the quarterly report for Acme Corp." }, ], maxTokens: 1024, temperature: 0.7, }); // result.text → "Acme Corp's Q1 2026 revenue was $2.3M, up 18% from Q4 2025..."

// Image generation const image = await ctx.ai.generateImage({ model: "sd3.5-large", prompt: "A professional infographic showing Q1 revenue growth", size: "1024x1024", quality: "standard", }); // image.url → "https://cdn.aeionos.com/ai-generated/..."

// Speech synthesis const audio = await ctx.ai.speak({ model: "eleven_flash_v2_5", input: result.text, voice: "rachel", speed: 1.0, }); // audio.audioUrl → "https://cdn.aeionos.com/ai-generated/..."

// 3D mesh generation const mesh = await ctx.ai.generateMesh3D({ model: "tripo3d", prompt: "A low-poly stylized tree with autumn leaves", }); // mesh.modelUrl → "https://cdn.aeionos.com/ai-generated/..." ```

Provider routing is automatic: Pass any model name and the registry dispatches to the correct adapter. No provider selection needed.

Usage tracking is automatic for all AI operations:

No configuration needed — it's event-driven. When any AI operation completes, ai.operation.completed fires and Aeion logs it:

```typescript // ai_usage_logs collection gets populated automatically // Every operation: timestamp, provider, model, tokens, cost, latency, status

// Query usage: const usage = await ctx.services["ai_usage_logs"].list({ where: { tenantId: { equals: tenantId }, createdAt: { gte: startDate, lte: endDate }, }, order: [{ field: "createdAt", direction: "desc" }], limit: 100, });

// Aggregate by provider: const byProvider = await usageService.getUsage({ tenantId, period: "monthly", groupBy: "provider", }); // → [{ providerId: "openai", operationCount: 4523, totalCost: 127.50 }, // { providerId: "anthropic", operationCount: 891, totalCost: 43.20 }] ```

Token counting:

```typescript // Pure JS, no API call — estimate tokens before making the call const tokens = estimateTokens(longText); // → 2847 tokens

// Pre-call cost estimation const cost = calculateTextCost("openai", "gpt-4o", tokens, 500); // → { inputCost: 0.014235, outputCost: 0.0075, totalCost: 0.021735 } ```

Cost caps prevent runaway bills from AI operations:

json // ai_budgets record (one per tenant + period) { "period": "monthly", "limitUsd": 500, "alertThresholdPercent": 80, "actionOnExceed": "soft-block" // warn | soft-block | hard-block }

How it works: Before every AI operation, Aeion checks the tenant's active budget row for that period against current spend. Crossing alertThresholdPercent fires an alert. If spend would exceed the limit, the configured actionOnExceed applies — warn (log + notify, call proceeds), soft-block (require admin override), or hard-block (reject the call).

Low-budget starter example:

json // ai_budgets record for a new tenant during their 14-day trial { "period": "monthly", "limitUsd": 5, "actionOnExceed": "hard-block" }

Navigate: AI → Dashboards → New Dashboard

json { "name": "Monthly AI Spend", "type": "cost", "config": { "metrics": ["totalCost", "inputCost", "outputCost"], "groupBy": "provider", "dateRange": { "start": "2026-05-01T00:00:00Z", "end": "2026-05-31T23:59:59Z" }, "chartType": "area" } }

Usage aggregations:

```typescript // By provider: which AI provider costs the most? groupBy: "provider";

// By user: which users are consuming the most AI budget? groupBy: "user";

// By operation type: text vs image vs audio vs 3D groupBy: "operationType";

// By model: which specific model is most expensive? groupBy: "model";

// By modality groupBy: "modality";

// By period period: "hourly" | "daily" | "weekly" | "monthly"; ```

Example: Top 10 users by AI spend:

typescript const topUsers = await usageService.getUsage({ tenantId, period: "monthly", groupBy: "user", limit: 10, order: [{ field: "totalCost", direction: "desc" }], });

No API calls, fully local, zero cost:

json { "providerId": "ollama", "name": "Ollama Local", "status": "active", "credentials": { "baseUrl": "http://localhost:11434" }, "defaults": { "textModel": "llama3", "embeddingModel": "nomic-embed-text" } }

Deploy on-premise:

```bash # Install Ollama on the server curl -fsSL https://ollama.com/install.sh | sh

# Pull models ollama pull llama3 ollama pull mistral ollama pull nomic-embed-text

# Aeion AI connects to localhost:11434 ```

Available Ollama models:

• Text: llama3, llama3.1, llama3.2, mistral, mixtral, gemma2, phi3, qwen2.5, codellama
• Embeddings: nomic-embed-text, mxbai-embed-large, all-minilm

For air-gapped environments (government, healthcare, finance): Ollama + Aeion AI gives you full AI capabilities without any data leaving the network.