01
Store once
AI provider keys (OpenAI, Anthropic, Google, Mistral, and more) are configured a single time by an admin in the Workbench. That is the last time anyone copies them by hand.
← Security
SECURITY
Stop distributing provider keys.
Store approved model credentials in the Workbench environment and grant users access through roles and sessions. Developers can use the model without copying the underlying secret into a notebook, repository, or chat thread.
The problem
Provider keys multiply faster than their owners remember.
The standard workflow is broken. You create an OpenAI key. You send it to a developer in Slack. It lands in a .env file. It gets committed to git. It ends up in three other places you will never find.
When the key needs rotating, you break everyone's local setup at once. When it leaks, you have no idea what accessed it or what data it touched.
Calliope AI Workbench centralizes the credential path. Admins configure approved provider access in the deployment environment, while users invoke it through an authorized session.
How it works
Credentials managed centrally. Access granted by policy.
02
Access by policy
Teams and users get access to credentials based on their role. No key distribution, no shared secrets in a password manager, no Slack threads.
03
Automatic credential injection
When a developer opens AI Lab, the IDE, or Chat Studio, credentials are injected into the session automatically. The tools just work, and nothing is written to disk.
04
No copying into code
Users do not need to paste a provider key into source files, notebook cells, or shared configuration to use the approved model integration.
05
Rotate in one place
Update a key in the Workbench and every tool that uses it picks up the new credential immediately. No developer action, no broken setups, no downtime.
06
Access records and optional policy
Workbench records workspace access. Add Zentinelle AI separately when the control model requires attributed model events, policy evaluation, enforcement, and evidence.
What you manage centrally.
AI Provider Keys
OpenAI, Anthropic, Google Gemini, Mistral, Cohere, and any other provider your team uses. Configure once, available everywhere in the Workbench.
Database Credentials
Connection strings for SQL, NoSQL, and vector stores used by AI Lab and DB Loadr. Developers connect to data sources without ever seeing the credentials.
Per-Team Scoping
Different teams reach different providers and models. Engineering gets Claude, analysts get GPT, sensitive workloads stay on local Ollama. All controlled from one place.
Bring your own keys
Already have provider contracts? Use them.
If your organization already has enterprise agreements with AI providers, store those credentials in the approved Workbench deployment and expose them only to the tools and roles that need them.
Keep the provider contract and token charges directly with the provider. Calliope AI Workbench adds central credential handling, while Zentinelle AI may be added for agent-level policy and evidence.
Stop distributing keys. Manage them centrally.
Give users access to approved model integrations without asking them to copy provider secrets into their daily tools.
