SCAIRE
LLM
Manage OpenAI-compatible LLM providers (the table below), per-call timeouts, and retry limits.
Provider Configuration
Each row is one OpenAI-compatible endpoint. The default provider is used by every agent unless an agent is assigned a specific provider on the Agents page. Changes take effect after an app restart (PHP-FPM reload / queue worker restart).
| Name | Driver | Model | Status | Actions |
|---|---|---|---|---|
|
Zai zai Default |
deepseek |
glm-5-turbo | Enabled |
|
Insights
Providers
1
1 enabled
Default
Zai
deepseek / glm-5-turbo
API Key
Set
on default provider
Chat Timeout
300s
per LLM step
Understanding Your LLM Configuration
Provider System
- Every agent uses the default provider unless overridden per-agent on the Agents page.
- Multiple providers let you switch models without reconfiguring every agent. Set a primary, keep a backup.
- Changes take effect after an app restart (PHP-FPM reload + queue worker restart).
Timeouts and Retries
- Per-call timeouts cap a single LLM step. If the provider is slow, the step fails and retries kick in.
- Whole-session ceilings cap an entire chat turn (all steps combined). Prevents a long agentic loop from running forever.
- Max retries handles transient network errors. Empty-response retries handles occasional blank replies from some models.
Cost Tracking and Drivers
Token Costs
- Per-model rates are tracked on the Token Pricing tab for deterministic cost reporting.
- Turn on Provider Token Reporting (Advanced tab) for exact usage numbers from the API, at a small per-request overhead.
Drivers
- The driver tells the SDK which API format to expect. Each provider can use a different driver depending on its compatibility.
- If a provider returns 404 or unexpected errors, try switching its driver. Some endpoints do not support every API path.