Provider Templates

A Provider Template is a named bundle of (provider, model, credential, default settings) that endpoints can reference. Without templates, each endpoint repeats the same provider config; with templates, you change one row and every endpoint inherits.

When to use templates

✅ You have many endpoints all hitting gpt-4o-mini with temperature=0.7 and the same OpenAI credential — make a template, attach all endpoints to it. ✅ You want to swap a model gateway-wide (gpt-4o-mini → gpt-4o) — edit the template, every endpoint follows. ✅ You want a single audit point for “what’s our default chat model” — the template name is it.

❌ You have one experimental endpoint with unique settings — manual mode is fine. ❌ Each endpoint genuinely uses a different provider+model — templates would just add indirection.

Anatomy

Field	Notes
`name`	Human-readable. Shown in endpoint pickers.
`slug`	Auto-generated URL-safe identifier.
`description`	Free text. Where you explain why this template exists.
`tags`	Free-form labels for filtering.
`provider_key`	Which adapter (`openai`, `anthropic`, …).
`provider_model`	Provider-specific model identifier.
`credential_id`	Which encrypted API key to use.
`temperature`	Default `0.0`–`2.0`.
`top_p`	Default `0.0`–`1.0`.
`max_output_tokens`	Default response cap.
`is_active`	Inactive templates are hidden from endpoint pickers.

Creating a template

Top-right user menu → Provider Templates → + New template.

The wizard has 4 tabs:

Tab 1 — Identity

Name: e.g. GPT-4o-mini Standard
Description: free text
Tags: comma-separated — appear as chips in the list

Tab 2 — Provider

Provider: dropdown of the 8 registered providers
Model: provider-specific identifier (e.g. gpt-4o-mini)
Credential: filtered by provider — only credentials that match show up

Tab 3 — Settings

Temperature (default 1.0)
Top P (default 1.0)
Max output tokens (default 1000)

Tab 4 — Activation

Single toggle: Active. Inactive templates can’t be attached to new endpoints (but existing endpoints keep working).

Save.

Using a template on an endpoint

In the endpoint wizard’s Tab 2 — Provider:

Pick Provider Template mode.
Choose your template from the dropdown.

The endpoint’s provider/model/credential are now read from the template at request time. The endpoint can still override runtime settings:

temperature (endpoint > template)
top_p (endpoint > template)
max_output_tokens (endpoint > template)

If the endpoint leaves these fields at defaults, the template’s values win. This pattern lets you keep “company-wide standards” in the template and “endpoint-specific tuning” on the endpoint.

What changes when I edit a template?

Editing a template immediately affects every endpoint that references it. The next request through any of those endpoints will use the new values. There’s no caching or per-endpoint snapshot.

So if you change the model from gpt-4o-mini to gpt-4o:

All endpoints with this template start hitting gpt-4o on the next request.
Their gateway_logs rows record provider_model = gpt-4o from that point on.
No restart, no migration, no downtime.

This is powerful — and dangerous. Always think “is this safe to change live?” before editing a widely-used template. Use deactivate to test the impact: deactivate a template, verify which endpoints fail-open or fail-closed, then iterate.

Deactivate vs delete

Action	Effect
Deactivate	Template is hidden from new-endpoint pickers. Existing endpoints with this template start failing requests. Reversible.
Delete	Removes the row. Endpoints with `provider_template_id` pointing to it have it nulled out (foreign key `nullOnDelete`). Those endpoints lose their provider config and break unless they had manual config too.

Examples

Cheap default

Name:       OpenAI Mini Standard
Provider:   OpenAI
Model:      gpt-4o-mini
Credential: OpenAI Production
Temperature: 0.7
Top P:       1.0
Max tokens:  1000

Use this for every endpoint that doesn’t need gpt-4o. When OpenAI releases the next mini model, edit one template.

Quality endpoint

Name:       Anthropic Sonnet Premium
Provider:   Anthropic
Model:      claude-sonnet-4-6-20251001
Credential: Anthropic Production
Temperature: 0.5
Top P:       1.0
Max tokens:  4000

Attach to endpoints where output quality matters more than cost.

Local fallback

Name:       Ollama Llama Local
Provider:   Ollama
Model:      llama3
Credential: Ollama Localhost (placeholder)
Temperature: 0.6

Use this in failover chains as a free local backup when paid providers are unreachable.

Detail page

Each template’s detail page shows:

Identity + provider config + settings
Endpoints using this template — list with link to each
Activate / deactivate / delete actions

Useful sanity check before editing: “what am I about to change?”

Next: Provider Settings — gateway-wide enable/disable.

© Akyros Labs LLC. All rights reserved.