Agent Vault in Action: Give Agents Access, Not Secrets
How to give AI agents access to the tools they need without handing them real credentials
- Date
- October 21
- Time
- 1 PM ET
- Length
- 30 minutes
AI agents need credentials to do anything useful. Teams give them API keys for GitHub, Linear, Datadog, and every other service the agent touches. Those keys are often long-lived, have full access, and get pasted straight into a config or .env file.
That creates real risk, because anything an agent can read, it can leak. A webpage with a hidden instruction can talk an agent into sending your keys to an attacker. Agents can write credentials into logs and send them to model provider servers. And once every team is spinning up its own agents, it gets hard to say which agent has access to what.
In this live demo, we'll walk through Agent Vault, Infisical's new product for securing AI agent infrastructure. You'll see how Agent Vault brokers credentials through a proxy, so your agents reach the services they need without ever reading a secret.
What you will learn
- Where agents can leak credentials, from prompt injection to logs and model providers
- How credential brokering gives agents API access without exposing real keys
- Connecting Claude Code, Codex, or your own agent to Agent Vault without code changes
- Defining which services an agent can reach with access bundles
- Setting time limits on agent access with expiring sessions
- Giving each agent its own identity and tracking its requests in session logs
Agenda
- 01Introduction
- 02Live Agent Vault demo
- 03Q&A
Who this is for
Platform, infrastructure, security, and engineering teams deploying AI agents that need access to production services, and anyone running untrusted code, like CI pipelines or third-party scripts, that needs credentials.
Speakers
- Jake HulbergDeveloper Relations (host)
- Kate NeuensMarketing Lead (moderator)
