# Gentle-AI: Persistent memory and deterministic workflow for AI coding agents > Gentle-AI is an open-source configuration framework for AI coding agents like Claude Code, Cursor, and Pi. It adds persistent memory (Engram), deterministic workflow (ODD), and review evidence (RDD) without agent lock-in, supporting 17 different agent platforms. **Gentleman-Programming/gentle-ai** — Gentle-AI configures the AI coding agents you already use: Claude Code, Cursor, OpenCode, Codex, Pi, and more. Choose persistent memory, Spec-Driven Development, curated skills, MCP servers, personas, and optional bounded review. Open source, no agent lock-in. - Repository: https://github.com/Gentleman-Programming/gentle-ai - Website: https://gentle-ai.gentlemanprogramming.com/ - Stars: 7,288 · Forks: 794 - Language: Go - License: MIT - Published: 2026-09-09 · Updated: 2026-09-09 · Language: en - Canonical page: https://hysenlabs.com/projects/gentleman-programming-gentle-ai ## Configuration for agents you already use, not a new agent Gentle-AI is designed to work with existing AI coding agents, not replace them. The README lists 17 supported integrations: Pi, OpenCode, Claude Code, Codex, Cursor, VS Code Copilot, Gemini CLI, Kilo Code, Kimi Code, Kiro IDE, Qwen Code, Hermes, Antigravity, Windsurf, OpenClaw, Trae, and Conductor. The project is written in Go and distributed as a standalone binary. You install it once and configure it globally or per repository. Each supported agent has its own configuration documented in docs/agents.md. There is no agent lock-in; you configure the agent you already use. ## Engram: Persistent memory between sessions Engram is Gentle-AI's persistent memory system. Instead of starting fresh each session, the agent writes down what it learns and reaches for that context before asking you the same questions again. The README states that the cost of a fresh session is not the tokens, it is you re-explaining decisions every morning. Engram stores project decisions, design patterns, build commands, and other context that the agent accumulates. On the next session, the agent checks Engram first, so context accumulates instead of resetting. ## ODD: Organic Driven Development workflow ODD (Organic Driven Development) is Gentle-AI's workflow abstraction. Small changes should not need a planning pipeline; large work should not lose context between sessions. ODD keeps small changes lightweight and gives substantial work a recoverable feature document. Under ODD, the agent explores code before changing it, checks the results, and keeps progress current so work can resume without rebuilding context. The README notes that the agent's guidance is proportional to the request: small fixes get small guidance, large refactors get more thorough planning. When Strict TDD is enabled, the agent captures a failing test before implementation, makes it pass, then refactors while tests stay green. When disabled, the agent still runs applicable functional checks. ## RDD: Receipt-Driven Development and bounded review RDD (Receipt-Driven Development) is enabled by default and opt-out. When the agent finishes a change, RDD freezes the candidate before review, so the evidence belongs to the exact version you are about to rely on, not to whatever the worktree looked like a moment later. The README explains that a model guesses the next step differently tomorrow and differently again for your teammate. That is the gap between a workflow and a suggestion. The gentle-ai binary owns native RDD review transitions. Review evidence is bound to the candidate rather than the model's recollection. You can disable RDD globally with gentle-ai review mode disable. Explicit choices remain OFF for that clone. ## Deterministic by design, not guessing Gentle-AI is built on the premise that a workflow is different from a suggestion. A model that guesses the next step guesses differently tomorrow. Gentle-AI owns the flow of work: which transitions are valid next, what evidence belongs to which decision, and how work resumes after interruption. The trigger rules are deterministic, not learned. You do not re-explain the same decisions because Gentle-AI keeps them in Engram. You do not wonder what the agent did because RDD freezes the evidence. You do not start over because ODD keeps progress recoverable. ## Configuration per repository and global Gentle-AI can be configured globally or per repository. Global configuration applies to all projects you work on. Clone-local configuration overrides global settings for a specific repository. Configuration uses TOML and YAML files in the docs/ and .claude/ directories. The go.mod file shows dependencies on charmbracelet/bubbletea (terminal UI), jedisct1/go-minisign (signature verification), and modernc.org/sqlite (local storage for Engram). The project uses bubbletea for an interactive TUI where you can configure settings and review progress. ## Integrations with 17 agent platforms Gentle-AI documents integrations with Pi, OpenCode, Claude Code, Cursor, and others in docs/agents.md. Each integration has different capabilities. The README notes that native configuration and feature comparison are documented there. Some agents support full ODD and RDD; others support subsets of capabilities. You choose the agent that fits your workflow and configure Gentle-AI to guide that agent. There is no Gentle-AI agent; it is entirely about configuring the agent you chose. ## Lightweight, deterministic, and evidence-focused Gentle-AI prioritizes three things: being lightweight enough to run on every session, keeping workflow deterministic so the next step is always knowable, and binding evidence (review checks, test results) to the exact code you are about to merge. It does not try to be an IDE or a model. It is a configuration layer that sits between you and your agent, holding context, enforcing workflow, and proving what happened. The binary is written in Go for speed and runs natively on macOS, Linux, and Windows. ## Conclusion Adopt Gentle-AI if you use an AI coding agent and want persistent memory across sessions, deterministic workflow guidance, and bounded code review without switching agents. Avoid it if you need a lightweight tool with minimal configuration. Before adopting, check that your agent is in the list of 17 supported platforms and that your team can adopt its workflow conventions (ODD, RDD, TDD). ## FAQ ### How do you install Gentle-AI? Download the latest binary from GitHub Releases or build from source. The README provides quickstart instructions at docs/quickstart.md. Configure it globally or per repository using TOML configuration files. ### How does Gentle-AI work with my AI agent? Gentle-AI configures the agent you already use (Claude Code, Cursor, Pi, or 14 others) by injecting workflow guidance, persistent memory, and review checks. It does not replace your agent; it enhances it. ### What is Engram? Engram is Gentle-AI's persistent memory system. It stores project decisions, design patterns, and context that your agent learns so that context accumulates across sessions instead of resetting. ## Sources - [Gentleman-Programming/gentle-ai on GitHub](https://github.com/Gentleman-Programming/gentle-ai) - [License: MIT](https://github.com/Gentleman-Programming/gentle-ai/blob/main/LICENSE) - [Project website](https://gentle-ai.gentlemanprogramming.com/) - [README](https://github.com/Gentleman-Programming/gentle-ai/blob/main/README.md) - [Releases](https://github.com/Gentleman-Programming/gentle-ai/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/gentleman-programming-gentle-ai