transient — A governance layer for accountable…

Pitch

Description

Comments

About this project

Transient provides the essential trust infrastructure for autonomous agents. It enables actions without human oversight while ensuring accountability through a robust governance layer. With tamper-evident receipts and independent verification, this project safeguards agent operations at scale without requiring modifications to the agent's code.

Transient provides a robust trust infrastructure for autonomous agents, focusing on governance and accountability in their operations. As these agents gain the ability to perform tasks independently, including writing code, making purchases, and executing APIs, it becomes crucial to ensure their actions are transparent and accountable.

Key Features

Governance Layer: Transient operates as a governance layer that functions outside the agent process, establishing an essential audit trail for all actions taken by the agent. It intercepts each subprocess call prior to execution, allowing for policy evaluation and ensuring that actions are permitted or denied according to pre-defined rules.
Tamper-Evident Receipts: Every action executed produces an immutable, Ed25519-signed receipt. This allows for independent verification of the agent's actions by any party without reliance on the issuing system, enhancing trust and accountability.
Comprehensive Memory and Verification: Transient indexes actions, patterns, and session context from its governance layer, rather than agent logs. This enables effective monitoring and verification of content-producing actions against declared intentions.

Installation

To use Transient, the transient-trace command-line interface (CLI) can be installed via pipx:

pipx install transient-trace

Usage

Transient is launched by wrapping the agent process, ensuring that all actions are governed and receipts are linked for comprehensive oversight. For example:

transient-trace run python agent.py
transient-trace run claude
transient-trace --mode strict --packages filesystem,code,privilege,shell run python agent.py

Operation Modes

Audit Mode: The default setting records all activities without blocking actions, useful for monitoring agent behavior.
Strict Mode: Enforces policy rules before action execution, ideal for production environments.
Permissive Mode: Logs policy violations without blocking, suitable for testing.

Intended Use Cases

Transient is designed for teams that need a governance framework for deploying autonomous agents, ensuring security, compliance, and accountability. It caters to organizations that require tamper-evident records and robust oversight of autonomous actions.

For more detailed information and technical documentation, please refer to the full documentation.