cyber.hendryadrian.com
Security News · helpnetsecurity.com

Tuskira Releases Open-Source Gateway to Protect AI Agent Credentials and Tool Access

· Original article ↗

Summary

Tuskira’s open-source AI Agent Gateway runs in the customer’s environment, checks permissions on tool calls, and supplies stored credentials. The article also flags risks from unbound profiles, stored prompt data, and permissive demo network settings.

Key points

  • The gateway mediates between AI agents, MCP tool servers, and model providers, running in the customer’s environment without a Tuskira account.
  • It checks tenant roles and profile permissions on each tool call; denied requests are logged and do not reach the backend.
  • For allowed calls, it retrieves credentials from encrypted storage so agents do not need to hold the underlying service tokens.
  • Keys without a profile binding let callers specify a profile in the request; Tuskira recommends binding each key to its intended profile.
  • The demo stack stores request and response bodies, which may contain prompts and code; storage can be disabled.
  • The shipped Docker Compose configuration allows connections to the host and loopback range for local testing; Tuskira advises removing those exceptions in shared deployments.

Article Details

Event Type
Release of Tuskira’s open-source AI Agent Gateway, which mediates AI-agent tool and model-provider requests and controls credential use.
Impact
The gateway can enforce tenant-role and profile permissions, keep backend credentials out of agent configurations, log calls, and restrict network access. The article also notes configuration risks: unbound keys may let callers select a profile, the demo stack stores request and response bodies by default, and the shipped Docker Compose configuration allows outbound connections to the host and loopback ranges.

Vendors

Amazon Web ServicesIt covers Anthropic directly or through AWS Bedrock, plus OpenAI and Gemini, and it records tokens and an estimated cost for each call.Anthropic#TuskiraGoogleOpenAI#TuskiraTuskiraTuskira’s AI Agent Gateway is an open-source gateway that enforces permissions between AI agents, MCP tool servers, and model providers while running in the customer’s own environment.

Products

AI Agent GatewayTuskira’s AI Agent Gateway is an open-source gateway that enforces permissions between AI agents, MCP tool servers, and model providers while running in the customer’s own environment.AWS BedrockIt covers Anthropic directly or through AWS Bedrock, plus OpenAI and Gemini, and it records tokens and an estimated cost for each call.Claude CodeThe project supports major AI tools and models, including Claude Code, Cursor, OpenAI, Gemini, and Anthropic.Codex CLIThe repository ships worked examples, covering Claude Code, Cursor, VS Code, Codex CLI, a Python agent and Kubernetes.Cursor#TuskiraDocker ComposeThe shipped Docker Compose file allows outbound connections to the host machine and the loopback range for local testing, and Tuskira’s instruction is to remove both on anything shared.Gemini#TuskiraGitHub#TuskiraJira#TuskiraKubernetesThe repository ships worked examples, covering Claude Code, Cursor, VS Code, Codex CLI, a Python agent and Kubernetes.VS CodeThe repository ships worked examples, covering Claude Code, Cursor, VS Code, Codex CLI, a Python agent and Kubernetes.WSL2The gateway runs on macOS and Linux, and Windows through WSL2 is untested.

Related Articles