SECRETS VAULT
Secrets vault for AI. Keys kept out of model context.
Pasting API keys into ChatGPT, Claude or agent chats does not scale. Once a key is in a prompt, it is in the model’s context, where it can be logged, kept in history and leaked. Joyst’s secrets vault stores secrets encrypted and supplies them through controlled substitution, which is designed to keep keys out of the model’s view.
BUILT FOR AI WORKFLOWS
Stop pasting API keys into AI chat
- Keys in chat history
Once a key is in ChatGPT, Claude or an agent chat, it is in the model’s context. It can be logged, kept in history and leaked in screenshots.
- Old habits fail audits
Teams share keys in Slack, drop them into system prompts and hope nobody copies them to the wrong channel. That fails audits.
- API key management for AI
Agents and tools need credentials at the right moment, without the model reading the raw secret. Pasting keys into chat cannot do that.
HOW IT WORKS
Controlled substitution
- The vault holds the key
The vault stores the secret. Your workflow refers to it by name, so the raw key is not written into prompts or agent definitions.
- Swapped in when needed
Joyst substitutes the value so the agent or tool can authenticate. It is designed to do this without adding the key to the prompt the model sees.
- Keys kept out of model context
Controlled substitution is how Joyst is designed to keep API keys out of the model’s context in AI workflows and agent setups.
PRIVATE BY DEFAULT
Encrypted vault, private by default
- Stored encrypted
Secrets are stored encrypted in the private vault, not in chat sidebars, Slack threads or prompts.
- Separate rights
Being able to use a secret in a workflow is separate from being able to view or manage it. Access follows the same rules as the rest of Joyst.
- Part of AI governance
Vault permissions sit within AI governance, so security teams can review permissions and the audit trail across the private catalogue.
IN PACKAGES
Secrets inside agents and MCP setups
- Referenced by name
Agents refer to secrets by name. Values are never pasted into the agent definition.
- MCP tool credentials
Connect MCP tools through the vault instead of embedding keys in chat or skill text.
- Prompts without keys
Keep credentials out of prompts. The vault holds the keys, and your prompts stay clean.
WHERE IT FITS
What Joyst is not.
- Not for infrastructure secrets
Joyst is not built for infrastructure secrets such as cluster credentials, database passwords and CI secrets. Keep those in AWS Secrets Manager.
- Alongside your existing tools
Joyst does not replace HashiCorp Vault for infrastructure. It sits alongside your existing secrets tools for AI workflows.
- Built for AI workflows
Joyst holds the secrets your agents and AI tools use, with controlled substitution. Infrastructure secrets stay where they are, and AI secrets stay out of chat.
RELATED
Keep exploring.
- AI governance
Permissions and an audit trail across the private catalogue.
Open - Prompt management
A team prompt library with versioning and sharing levels.
Open - MCP catalogue
Connect, document and approve MCP servers.
Open - Skills registry
Manage, version and move skills across AI tools.
Open - Stop pasting API keys
Why pasting keys into AI chat fails audits.
Open - Controlled substitution
Use vault secrets by name. Designed to keep keys out of the model’s context.
Open - Add and use secrets
Store keys in the vault and refer to them safely.
Open - OAuth for MCP servers
How MCP OAuth sign-in works, step by step.
Open - Secrets management tools for AI agents
How a vault for AI agents differs from infrastructure tools.
Open - Security for AI agents
Controls to put in place before an agent runs.
Open
Secrets vault FAQ
Do not paste keys into prompts or skills. Store the value in the vault and refer to it by name. Joyst does not strip secrets out of that text. Substitution happens only when Joyst makes an outbound HTTP call. The value is placed in that request on the server, then removed from the response. It is not swapped into the prompt the model reads.
Joyst is a secrets vault for AI workflows. It does not replace AWS Secrets Manager or HashiCorp Vault for infrastructure secrets.
The agent package stores the secret’s name, not its value. That does not strip a key out of prompt or skill text. Joyst substitutes the value only on an outbound HTTP call it makes, and it removes the value from the response.
Anything in a prompt can enter the model’s context, its history and its logs. Store keys in the vault instead.
Secrets are one of the five types of item covered by permissions and the audit trail in the private catalogue.
Secrets are private by default. You cannot read a secret’s value back. A list or a description never returns the value. Only the owner can update or delete a secret. Organisation admins can use secrets. Members have restricted access. Other access is set through user groups.
You cannot read the value back. To change it, overwrite it with a new value. Secrets do not expire.
Keep the value in the vault. An agent package stores the secret’s name, not the value. An MCP connection stores the endpoint, not the credentials. Joyst does not pass live calls on to third-party MCP servers. When Joyst makes an outbound HTTP call, it can use a vault secret in that request.
Keep API keys out of model context
Create an organisation account and bring skills, prompts, MCP servers, secrets and agents into one private catalogue.
- Private to your organisation
- Draft, then publish
- Limited open beta