HOW IT WORKS
Controlled substitution for AI agent authentication
AI agent authentication often means pasting keys into prompts, where they end up in model context. Controlled substitution keeps the secret in the vault and inserts it on the server when it is used. It is designed so that the raw value does not appear in the prompt the model sees.
THE IDEA
Reference in, value out of band
- The problem
Pasted keys sit in prompts, provider logs and chat history.
- The approach
An agent or tool asks Joyst to use a named secret. Joyst substitutes the value where it is needed.
- The aim
The aim is to keep keys out of the model’s context.
SCOPE
Be precise
- The design aim
Controlled substitution is designed to keep raw secret values out of the prompt the model sees.
- Handled on the server
Joyst performs the substitution on its own server. The model is never asked to remember a key.
- Secrets only
Joyst does not route all MCP traffic. Substitution applies to secrets only.
- Details can change
Where substitution is available can change between releases.
IN PRACTICE
Controlled substitution. From vault to call.
- Store
The secret is stored encrypted in the vault.
- Permission
The caller must have permission to use the secret.
- Request
The tool refers to the secret by name, for example in an authorised HTTP helper.
- Substitute
Joyst applies the value on the server. This is designed to keep the key out of the prompt.
RELATED
Keep exploring.
- Resources hub
Practical guides to setting up and using Joyst.
Open - Secrets vault
Keys kept out of model context.
Open - Add and use secrets
Store keys in the vault and refer to them safely.
Open - How to stop pasting API keys
Store keys in the vault instead of pasting them into chat.
Open - Create an agent package
Assemble skills, MCP servers, prompts and secrets into one setup.
Open - Agents
Package skills, prompts, MCP servers and secrets into reusable setups.
Open - Stop pasting API keys: why it fails
Why pasting keys into AI chat fails audits.
Open - MCP authentication
How MCP OAuth sign-in works, step by step.
Open - Secrets management for AI agents
How a vault for AI agents differs from infrastructure tools.
Open - AI agent security
Controls to put in place before an agent runs.
Open
Controlled substitution FAQ
No. Environment variables can still end up in prompts. Substitution is designed to keep the value inside the vault.
It should not. Tools that need to authenticate should receive credentials separately from the prompt.
Keep keys in the vault. Substitute them at use time.
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