LLM Providers
Configure and manage AI service providers that power your ExecAssist agents.
Overview
LLM (Large Language Model) Providers are the AI services that power agent intelligence. ExecAssist supports multiple providers, allowing you to choose based on capability, cost, and compliance requirements.
You connect and manage providers on the AI Models page. Connecting a provider takes one thing: its API key. There is no separate provider page and no list of models to tick. Once a provider is connected, every approved model it offers can be chosen for a job.
If you have an old link or bookmark to a provider's settings page, it still works. It now opens the AI Models page.
Supported Providers
| Provider | Models | Best For |
|---|---|---|
| Anthropic | Claude Sonnet, Claude Opus, Claude Haiku | General use, excellent reasoning |
| OpenAI | GPT-4, GPT-4 Turbo, GPT-3.5 | Wide compatibility, proven track record |
| Gemini Pro, Gemini Flash | Multimodal capabilities | |
| xAI | Grok | Alternative reasoning models |
| Azure OpenAI | GPT-4, GPT-4 Turbo | Enterprise compliance, data residency |
Hugging Face and OpenRouter can be connected too. They give you open models and models from many makers through one key.
Adding a Provider
Go to Workforce > Intelligence > AI Models. There are three places to connect a provider, and they all use the same short form.
From Choose a Setup
The first time you open the page, or after you click Change setup, you see the setup cards.
- A setup card (Anthropic, OpenAI, xAI): click Set up [provider name]. The first step of the drawer, Connect, asks for the key.
- A provider card (Google Gemini, Hugging Face, OpenRouter): click Connect [provider name]. A small window asks for the key. These providers have no ready-made lineup, so after connecting you choose a model for each job yourself.
- Microsoft Foundry setups connect through your own Azure instead of a key. Click Set up your Azure and follow the steps.
From Your Setup
At the bottom of Your setup, the Connections strip has an Add a connection button. Pick Anthropic, OpenAI or xAI and a small window asks for the key. Picking Azure AI Foundry opens its Azure setup page.
From the Model Picker
When you click a job's model, the picker also lists models from providers you have not connected, under Needs a connection. Select one and the details on the right show a key field. Once the key works, the model stays selected and you can use it straight away.
Entering the Key
- For Anthropic, OpenAI and xAI, click Where do I find this? if you need the steps for getting a key
- Paste the key into the field
- Click Check key
If the key works you see "Key works. Every [provider] model is now in your catalog." If it does not, you see "That key did not work. Check you copied all of it." Nothing is saved until a key works.
No Models to Add
You do not add models one at a time any more. Connecting a provider makes every approved model it offers available in the model picker. For Hugging Face and OpenRouter, the picker shows the recommended models.
Managing Providers
View Connected Providers
Your setup's Connections strip shows one tile per connected provider, with its logo, its status ("Working" or "Key not working") and when it last succeeded.
Replace a Key
- On the provider's tile, click Manage
- Under Replace the key, paste the new key and click Check key
The key in use stays in place until the new one works, so your agents are not interrupted.
Disconnect a Provider
- On the provider's tile, click Manage
- Under Disconnect, click Disconnect [provider name], then Yes, disconnect
Your agents stop using that provider. An agent that was set to one of its models goes back to your setup's model. You can connect the provider again at any time with a key.
Note: You cannot disconnect a provider while one of your nine jobs still uses it. The window names the jobs, for example "Writes uses Hugging Face. Choose another model for that job first." Change those jobs to another model, then disconnect.
Outermind Credit and Microsoft Foundry are managed on their own pages. Their Manage link takes you there.
Provider Status
Healthy (Green)
The provider is connected and responding normally.
Unhealthy (Red)
The provider cannot be reached. Common causes:
- Invalid or expired API key
- Network connectivity issues
- Provider service outage
- Rate limiting
Checking Provider Health
- Go to Workforce > Intelligence > AI Models and open Your setup
- A provider whose stored key is failing shows its Connections strip tile in amber with "Key not working", and each job row on that provider shows a "Provider key not working" chip
- Click Manage on the tile to see its status and replace the key
Cost Management
Understanding Costs
LLM costs depend on:
- Input tokens - Text sent to the model
- Output tokens - Text generated by the model
- Model choice - More capable models cost more
Monitoring Usage
View costs in Governance > Activity History > Platform Usage:
- Total spend by provider
- Cost per execution
- Token usage trends
- Per-agent cost breakdown
Cost Optimization Tips
- Use appropriate models - Haiku for simple tasks, Sonnet for complex
- Optimize prompts - Shorter, clearer instructions use fewer tokens
- Set token limits - Configure max tokens per response
- Monitor outliers - Check for unexpectedly expensive executions
Best Practices
Multiple Providers
Consider configuring multiple providers for:
- Redundancy - Failover if one provider has issues
- Cost optimization - Use cheaper providers for simple tasks
- Compliance - Azure for data residency requirements
- Capability matching - Different models for different use cases
Security
- Rotate API keys regularly
- Use environment-specific providers (dev, staging, prod)
- Monitor for unusual usage patterns
- Limit access to provider management
Provider Selection
| Scenario | Recommended Provider |
|---|---|
| General use | Anthropic (Claude Sonnet) |
| Cost-sensitive | Anthropic (Claude Haiku) or GPT-3.5 |
| Enterprise/compliance | Azure OpenAI |
| Complex reasoning | Anthropic (Claude Opus) or GPT-4 |
Troubleshooting
Invalid API Key
Symptoms: Provider shows "Key not working", agents fail
Solutions:
- Verify the API key is correct (no extra spaces)
- Check the key hasn't expired
- Ensure the key has appropriate permissions
- Replace it: on Your setup, click Manage on the provider's tile and use Replace the key
Rate Limiting
Symptoms: Intermittent failures, 429 errors
Solutions:
- Check your provider's rate limits
- Reduce concurrent agent executions
- Upgrade your provider plan if needed
Connection Timeout
Symptoms: Provider occasionally shows unhealthy
Solutions:
- Check network connectivity
- Verify firewall rules allow outbound HTTPS
- Check provider status page for outages