Models & BYOK
Every agent action is powered by a large language model. The Models tab (in Settings) lets you pick which model your workspace uses, set an automatic fallback, and — if you prefer — bring your own key (BYOK) so AI usage bills directly to your provider accounts instead of Syntrum's.
You reach it from the sidebar under Admin → .
Model settings are workspace-wide and are an Admin-and-above control, alongside the other governance settings. Operators and Viewers can't change model routing.
Model routing#
Two choices define how your agents get their intelligence:
- Default model — every agent action runs on this model. Pick it from the dropdown; the provider is shown next to each option.
- Backoff model — an automatic fallback used when the default is rate-limited or temporarily down, so your workforce keeps running. Choose one, or select None — no fallback.
The backoff model can never be the same as the default. If you set the default to a model that's currently the backoff, the backoff clears automatically.
Supported providers#
The model list spans the major providers, including:
| Provider | Example models |
|---|---|
| Anthropic | Claude Sonnet, Claude Haiku |
| OpenAI | GPT-5, GPT-5 mini |
| Google AI | Gemini 2.5 Pro, Gemini 2.5 Flash |
| xAI | Grok |
| Mistral | Mistral Large |
| DeepSeek | DeepSeek V3 |
BYOK — Bring your own key#
By default, agents use Syntrum's own API key and usage is billed through your plan. With BYOK, you paste in your own provider keys and agents bill directly to your accounts with that provider.
To add a key:
- In the BYOK — Bring your own key section, find the provider.
- (Optional) Pick which model that provider should use.
- Paste the API key into the field and click Save.
Once saved, the row shows the key's last four digits and when it was added (e.g. ••••a1b2 · added Jul 9, 2026). To stop using it, click Remove — agents fall back to Syntrum's key.
Key safety#
Your provider keys are stored server-side and encrypted. Only the last four digits are ever shown back to you — the full key is never displayed again after you save it. Removing a key deletes it.
How routing resolves#
When an agent is about to act, Syntrum resolves which model and key to use through a cascade — workspace → agent → tool — using the most specific setting available and falling back to the workspace default. In practice: your workspace default (and BYOK keys) apply everywhere unless something more specific is configured, with the backoff model catching provider outages.
The Models tab is pre-wired: if the backend model-settings endpoint hasn't shipped in your environment yet, your choices are saved to this browser and an amber note says so. They start syncing server-wide automatically once the endpoint is live.
Cost visibility#
Your AI spend — total cost, tokens, and calls, broken down by model — appears in the AI usage cost section of Reports. Per-agent cost attribution isn't available yet because all agents share one key; bringing your own keys routes spend to your own provider dashboards for finer accounting.
What's next#
- Reports — see what your model usage costs.
- Workspace settings — the other tabs in Settings.
- The three planes — how the workspace→agent→tool cascade governs behaviour, not just models.