Add your own LLM key
Add a provider key to a BYOK pool, scope the pool to your pipelines, set a monthly cap, and keep keys rotated — so inference runs on your account.
Adding your own LLM key routes IDPForge's model calls through your provider account. Keys live in provider pools under Settings → Integrations.
Workspace admins. Adding and rotating provider keys is an admin action.

The BYOK tab under Settings → Integrations. Each pool holds one provider's keys; add a key to a pool, or add a new pool with Add.
1. Open the BYOK tab
Go to Settings → Integrations and open the BYOK (LLM) tab. Existing provider pools are listed with their key count, region, and monthly cap.
2. Add a key to a pool
Choose Add key on the pool for the provider you want, or Add to create a new pool first. Give the key an alias you will recognise and paste the provider credential. The secret is stored encrypted; only its alias and status are shown afterward.
3. Test the connection
After adding the key, use the pool's Manage controls to confirm the key reaches the provider before you route real traffic through it. A key that cannot reach its provider should be fixed or disabled, not left in a routing pool.
4. Set the pool's monthly cap
Give the pool a monthly spend cap so BYOK inference cannot run away — the cap is shown on the pool row (for example $5,000/mo cap). A cap of $0 leaves it uncapped.
5. Choose how pipelines use the pool
Pipelines route through a provider's default pool automatically. Set a pool as default with Set default, or pin a specific pool on a pipeline only when you need to isolate its rate limits.
6. Rotate and retire keys
Rotate keys on a schedule — the pool row shows when each was last rotated. Disable a key to take it out of rotation while keeping its history; delete it to retire it for good.
Edge cases
- A configured provider with no usable key fails the job. If a pipeline is set to use your key for a provider but every key is disabled or unreachable, the job fails closed rather than falling back to managed inference — see Bring your own key.
- Sandbox and production keys are separate. Keys are partitioned by environment; add production keys from a production context.
- The org routing lock overrides pools. While the lock is on, every pipeline runs on IDPForge-managed inference regardless of pool configuration.