Workspace admin

Use your own model keys

Connect your own LLMGateway, OpenAI, Anthropic, or OpenRouter key so Veriom's agents run on your account and use no processing units.

This page is for a workspace Admin who wants Veriom's AI agents to run on the company's own model account. By default every agent runs on Veriom's platform models and uses processing units. With your own key, the provider bills you and Veriom uses no processing units for those calls.

Before you start

  • You have the Admin or Owner role.
  • You have an API key from one of these providers:
ProviderKey looks likeNotes
LLMGatewayllmgtwy_...Development keys (llmgdev_...) are refused in production
OpenAIsk-proj-... or sk-svcacct-...Admin keys (sk-admin-...) are refused. An EU endpoint is available
Anthropicsk-ant-...Admin keys are refused
OpenRoutersk-or-...
OpenAI-compatible endpointAnyEnterprise plan only. Needs an HTTPS address

Connect a key

  1. Open Settings and select AI models.
  2. Click Connect and paste the key. Veriom recognizes the provider from the key's shape.
  3. Click Check and save and confirm your identity.

Veriom checks the key with the provider and lists its models. The key is encrypted when saved and never shown again.

Route agents to your models

Veriom groups its agents into reasoning tiers. Under Model routing, choose a ready model from your key for each tier, then click Activate routing policy.

A model must be qualified by Veriom before agents can use it. Models still waiting show Awaiting operator qualification. Click Request qualification to ask for one.

Tiers you do not route keep running on Veriom's platform models.

Veriom's platform models

Veriom names its platform models by the work they do. Model settings, audit detail and the activity log show these names:

Veriom modelWhat it does
Veriom/Security-LiteFast work: routing questions, formatting answers and checking cited evidence
Veriom/Security-IDeep reasoning: security review, impact analysis, remediation and Ask
Veriom/Security-MaxThe highest-depth tier, reserved for explicitly approved decisions. No agent runs on it by default
Veriom/Architect-IArchitecture work: rebuilding components, relationships and trust boundaries, and writing the architecture narrative
Veriom/Index-IEvidence indexing for code and evidence search

Models on your own key keep the names your provider gives them in model settings and audit detail. The activity log can show the Veriom name instead for some agent runs.

What changes when you use your own key

  • The provider bills you for those calls. They use no processing units.
  • OpenAI calls are sent with storage turned off (store=false). OpenRouter calls are sent with data collection denied.
  • If your workspace has a spend cap for your own keys, agent work on them stops with customer_spend_cap_reached once the cap is reached.

Replace or remove a key

To replace a key, paste the new one with Paste the new key on the connected key. To remove it, revoke it from the list of connected keys. Tiers that used it need another model.

If something goes wrong

CodeWhat it meansFix
invalid_keyThe provider rejected the keyCopy the key again from the provider
admin_key_rejectedIt is an admin keyCreate a normal project or service key
insufficient_permissionsThe key cannot list or call modelsGive the key model access in the provider's console
quota_exhaustedThe provider account has no credit leftAdd credit with the provider
provider_unreachableThe provider did not answerTry again in 30 seconds
plan_not_allowedThis provider type is not on your planOpenAI-compatible endpoints need Enterprise

Was this guide useful?

Your response stays in this browser unless you open a GitHub issue.
View source