Torrix
GitHubStart free

Torrix for agents · Pro (up to 10 agents) and Enterprise

Authority and evidence for AI agents.

Give every agent its own identity, let people delegate scoped authority, and check each tool call that goes through the Torrix gateway against policy before it reaches your systems.

See how it works

Problem

Agents act for people. Most stacks cannot prove which person, or under which rules.

When an agent moves money or changes a record, you need one answer you can show an auditor.

  • API keys are shared.
  • Logs are scattered.
  • Approvals happen in chat.

How it works

Every tool and MCP call through the Torrix gateway is checked against who the agent acts for, its delegated scope and your policy.

Capabilities

Everything between a person, an agent and a tool.

hashed · scoped · revocable

Identity

Each agent gets its own trxa_ credential. Import agents built in Bedrock, Copilot Studio or your own code.

Signed

Delegation

People grant agents scoped authority with limits and expiry. Chains can only narrow, never widen. Grants are signed.

Deny overrides

Policy

Deny-overrides rules with risk scoring return allow, deny or require approval. Every decision is stored before the call runs.

Single-use

Approvals

Single-use approvals bound to the exact action and arguments. Alerts go to Slack, a webhook, PagerDuty or email.

Gateway

The MCP gateway filters tool discovery to each agent’s scope and checks every tools/call that passes through it. Agent-to-agent delegation is available on Enterprise.

# Point the agent’s MCP client at Torrix, not at your tools
{
  "url": "https://torrix.internal/gateway/mcp/erp-mcp",
  "headers": {
    "authorization": "Bearer trxa_AGENT_CREDENTIAL",
    "x-torrix-approval-wait": "120"
  }
}
# Every tools/call through this URL is checked, held or denied first

Start

Pick one workflow. We will govern it with you.

Tell us which agents you run and which systems they touch. We will set up Torrix with you on your own infrastructure.

Compare editions