gentle-ai: Add Memory and Workflows to Coding Agents

Overview

Gentle-AI is a Go-based tool built around the native configuration of AI coding agents you already use, including Claude Code, Cursor, OpenCode, Codex and Pi. It aims to address lost project context and inconsistent agent workflows by adding persistent memory, a structured but proportional development process, and review evidence. It supports macOS, Linux and Windows, and configures agents rather than installing them.

The shared workflow is not identical across integrations: available features, including delegation and review, depend on each agent's capabilities. Gentle-AI is for developers who want a repeatable way to guide AI-assisted work while keeping their existing agent choice.

Key Features

• Configures 17 coding-agent integrations using their native configuration.
• Engram persistent memory helps agents retain project context between sessions.
• Organic-Driven Development (ODD) keeps small work lightweight and records substantial authorized work for recovery across sessions.
• Receipt-Driven Development (RDD) binds review evidence to a specific candidate and adjusts review depth to assessed risk; it can be disabled.
• Includes curated skills, optional personas, Context7 MCP documentation access, and a read-only CodeGraph symbol graph.
• Snapshots agent configuration before writes and provides a read-only gentle-ai doctor health report.
• Includes a security deny-list for credential files and directories such as ~/.ssh and .env.

Use Cases

• A developer who repeatedly re-explains project decisions can use Engram to preserve context between agent sessions.
• A team using different coding agents can apply a shared workflow without requiring everyone to switch tools.
• A developer handling both small fixes and substantial features can keep straightforward changes inline while preserving plans and progress for work that needs to resume later.
• A maintainer who wants review evidence tied to the exact change can use RDD, while retaining control of commits, pushes and releases.

Getting Started

On macOS or Linux, install with:

curl -fsSL https://raw.githubusercontent.com/Gentleman-Programming/gentle-ai/main/scripts/install.sh | bash

Then run gentle-ai to choose agents and components, and gentle-ai doctor to check the setup. See the README and Quickstart for platform-specific prerequisites and details.

Alternatives

• deja-vu: Deja-vu focuses on searching existing coding-agent histories and recalling them through MCP and hooks, rather than configuring agents with broader workflows and review.
• ai-memory: ai-memory captures agent activity in a git-backed Markdown wiki and supports handoffs, rather than providing a multi-agent configuration and development workflow layer.
• plur: PLUR provides shared memory as local YAML files, while gentle-ai also configures agents with development workflows and evidence-based review.
• cq: cq centers on querying and validating reusable lessons from past work, rather than setting up persistent memory and workflows across coding agents.

Considerations

The project is active, with regular commits and releases, and has contributions from many people; measured contributor activity is not concentrated in one person. Its issue and pull-request handling is active as well. These signals indicate ongoing development, but do not guarantee that every agent integration has feature parity.

Gentle-AI configures existing agents rather than supplying them, and support varies by integration. The repository also distinguishes Gentle Shell from Gentle-AI itself: installing or updating Gentle-AI does not establish Pi runtime parity. The MIT license applies to the code; the README notes that the Gentle AI and Engram names and logos are trademarks, so the license does not grant permission to imply endorsement or official affiliation.