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, Organic-Driven Development, curated skills, MCP servers, personas, and optional bounded review. Open source, no agent lock-in.
★ 9,214⑂ 899👁 46⑃ 2,508● 3,561▤ 298
+41
日增长
9,214
stars
899
forks
46
watchers
2,508
prs
3,561
issues
298
releases
README
Gentle-AI™
The deterministic engineering environment for the AI agent you already use.
Website •
Quickstart •
Docs •
Wiki
Your agent writes code, then forgets everything. It has no opinion about your project,
and no way to prove what it did beyond asking you to read every line.
Gentle-AI gives it memory, a workflow, and evidence.
See it in action
One prompt, from idea to reviewed commit: memory, workflow, and evidence in a real session.
https://github.com/user-attachments/assets/fa5c0cfe-06e7-4c0d-bd6e-8ac7cb934339
Prefer Spanish subtitles?
https://github.com/user-attachments/assets/6d2bc422-a4dd-4ecf-a04b-fcd3bea7fea9
--
If Gentle-AI made your agent worth trusting, a star helps other people find it.
WORKS WITH THE AGENT YOU ALREADY HAVE
Pi ·
OpenCode ·
Claude Code ·
Codex ·
Cursor ·
VS Code Copilot ·
Gemini CLI ·
Kilo Code
Kimi Code ·
Kiro IDE ·
Qwen Code ·
Hermes ·
Antigravity ·
Windsurf ·
OpenClaw ·
Trae ·
Conductor
17 integrations · native configuration · compare capabilities →
Support the project
We deliberately don't depend on investors: no outside agenda deciding the roadmap and no features locked behind paywalls.
Donations through GitHub Sponsors go straight into maintenance, new features and keeping everything free and ours.
Can't donate? A ⭐, a bug report or a pull request helps too.
Sponsorship never buys rankings, default selections or endorsements; see TRADEMARKS.md.
Features
---
Engram™ — Keep your project context
The cost of a fresh session is not the tokens — it is you, re-explaining the same decisions every morning. Engram removes that: your agent writes down what it learns as it goes and reaches for it before it reaches for you, so context accumulates instead of resetting.
Docs →
---
ODD — Keep small work small
Small changes should not need a planning pipeline, and larger work should not lose its context between sessions. Organic Driven Development (ODD) keeps understood changes lightweight and gives substantial, authorized work one recoverable feature document. Questions, reviews and proposals stay read-only until you ask for a change. The agent explores before changing code, checks the results, and keeps progress current so work can resume without rebuilding the plan.
Tasks are sized by understanding, risk, and resumability, never by file count. A small task — understood, with contained risk, and resumable from the diff alone — runs inline: no subagents, no task document. Work becomes large only when it can't be resumed from the diff — several sessions, external waits, or requirements too long to survive compaction — never because of its size. The feature document of a large task holds your spec verbatim, and every handoff points at it instead of paraphrasing it.
https://github.com/user-attachments/assets/857a2195-34b8-4720-87af-f71bca3dec92
Docs →
---
Delegate for a reason — Not for Top issues
#4571bug(review 2.8.2): correction_required capture's own status_continuation classifies the just-minted authority corrupted_or_unverifiable_authority, so the bounded correction is unreachable395 comments · 2026-09-14#5422[Automated provider defect] bound STATUS after granted START times out pre_native for ~800+ changed-line candidates on 4.0.0 stable (Windows / Claude Code) — follow-up to #5094159 comments · 2026-10-09#4655bug(review): bound STATUS still times out pre_native after a granted START on a 3-path medium candidate (2.9.1 stable) — regression of #3733 / #4258109 comments · 2026-09-16#5129Review integration blocks on undocumented intended-untracked capture66 comments · 2026-09-30#5209[Automated provider defect] Claude Code: a granted, never-reoffered START's first capture-result lands directly on correction_required, and the bound STATUS goes terminal corrupted_or_unverifiable_authority with zero admitted captures (3.7.0)58 comments · 2026-10-02#5055bug(review): lens_context_inspection_failed on every fresh Claude Code slot for a candidate with a ~2.6 MB binary-attributed path; refresh-and-retry reoffers the same failing slots (3.7.0)51 comments · 2026-09-27#4790bug(review): the frozen-evidence budget is validated only after consent and lineage creation — validate at inspect/START and recommend a commit-range slice49 comments · 2026-09-19#2029fix(review): relay admission feedback and require canonical proof paths38 comments · 2026-07-30
Discussions
#6117Solicitud para colaborar como maintainer/revisor de la comunidad0 comments · 2026-11-13#6052DeepSeek harness0 comments · 2026-11-10#6043Supported budget-preserving disposition after initial unknown_causality escalation (CLI 4.3.1)1 comments · 2026-11-08#6016Supported disposition after a Pi verifier exits with EDQUOT before agent_settled0 comments · 2026-11-07#5978noob: cómo _uso_ Gentle AI?1 comments · 2026-11-05#5846Proposal: configurable ODD evaluation with automatic execution graphs0 comments · 2026-10-29#5767pi-reply-thread — keep track of the replies one prompt produces and the questions still waiting on you (Pi / Gentle Shell, MIT)1 comments · 2026-10-26#5764PRoblema en la instalacion0 comments · 2026-10-25
Latest release
v4.5.0 — Stopped Reviews Report Their Findings, Windows 10 22H2 Installs v4.5.0 2026-11-13Gentle AI v4.5.0 is a minor release over v4.4.0. A review that stops at correction_required or escalated now returns the findings it admitted as non-blocking to a client that explicitly opts in, and a correction-plan request that cannot be built reports correction_request_unavailable instead of a corrupted authority. The Separate installer qualifies Windows 10 22H2 (build 19045) on amd64, Codex receives the core-plus-on-demand module split, and Gentle Shell stops refusing a normally used Windows installation on runtime-state byte counts while keeping one recent replacement backup. One verifica