AI Provider Management (Admin)
AI Provider Management (Admin)
This article is for Kinexa administrators who manage the AI infrastructure at master.kinexa.id/setting/ai-provider.
Accessing AI Provider Settings
- Login to master.kinexa.id as an Admin or Super Admin
- Go to Settings > AI Provider in the sidebar
- You will see all configured AI providers and their status
Managing Providers
For each provider, you can:
- Enable/Disable — Toggle the provider on or off
- Health Check — Click the health check button to verify the provider is responding
- Edit Endpoint — Configure the API endpoint URL
- Set API Key — Enter or update the provider's API key
- Test Connection — Send a test request to verify everything works
Tier AI Recommendations
Use this table when configuring which AI model serves each pricing tier:
| Tier | Recommended Model | Provider | Reasoning |
|---|---|---|---|
| Free | Qwen 2.5 7B | Ollama | Zero cost, good quality |
| Starter | Qwen 2.5 7B | Ollama | Zero cost, good quality |
| Pro | Claude Haiku | Anthropic | Fast, cost-effective cloud AI |
| Business | Claude Sonnet | Anthropic | High quality, balanced cost |
| Enterprise | Claude Opus | Anthropic | Highest quality, complex reasoning |
Server Requirements for Ollama
If you are running Ollama on your own server:
| Model Size | Minimum RAM | Recommended RAM | CPU Cores |
|---|---|---|---|
| 3B (Phi-3) | 4 GB | 8 GB | 2+ |
| 7-9B (Qwen, Llama, Mistral) | 8 GB | 16 GB | 4+ |
| 13B | 16 GB | 32 GB | 4+ |
| 70B+ | 64 GB | 128 GB | 8+ |
GPU acceleration is recommended but not required. Ollama supports NVIDIA CUDA and Apple Metal.
Health Check Indicators
- Green — Provider is healthy and responding
- Yellow — Provider is slow (response time > 5 seconds)
- Red — Provider is unreachable or returning errors
Best Practices
- Always have at least two providers enabled for fallback
- Run health checks daily or set up automated monitoring
- Keep Ollama as a fallback even if you primarily use cloud providers
- Monitor API costs in the provider dashboard