Delegation, explained · 3:33
The whole delegation policy in one sitting: inline by default, the named reasons to hand work off, and verification in proportion to risk.
The Gentle stack for engineering teams
Turn the AI coding agents you already pay for into disciplined engineers.
Gentle-AI is the best practices of the market, and a few beyond it, packed into the agents your team already uses: Claude Code, Cursor, OpenCode, Codex, Pi and a dozen more. You ask in plain language; it picks the right route, remembers, verifies and proves its work. Open source, one binary per machine, no agent lock-in.
- 1
- command to install
- 15+
- agent runtimes
- 5
- open-source tools
Works with Claude Code · OpenCode · Cursor · Codex · Gemini CLI · VS Code Copilot · Pi · Windsurf · and 7 more
In one minute
What Gentle AI is, in under a minute.
What changes
Every session starts from zero
Engram keeps decisions, conventions and bug fixes, so the next session, and every other agent, starts where the last one stopped.
Code before requirements
The agent explores the code first and asks one focused question when a real decision is open, before it writes a line.
One request becomes a wall of changes, tested late or never
Test-first changes in small commits sized for a human to read, with tests and docs travelling with the code.
Nothing to memorise. Ask for the outcome; the agent finds the route.
- "add retries to the upload handler"
- "why does login flake"
- "plan the billing rewrite"
How it works
Four layers, each built on the one below.
Four short episodes, about a minute each. Watch them in order: the base that coordinates, the workflow it follows, the helpers it delegates to, and the review that looks at the result from outside.
Layer 1 · the base · 0:48
el Gentleman
You give the intent; it coordinates, asks only about real decisions and reports honestly. Push, PR and merge stay yours.
Layer 2 · the workflow · 1:04
ODD
Seven steps from authorize to close. Small work runs inline; large work gets a feature document in your exact words and small commits.
Layer 3 · delegation · 1:08
Focused agents
A read-only explorer, a writer that never commits and a read-only verifier, used only for a named reason, never by file count.
Layer 4 · the outside view · 0:58
Native review
It freezes the exact change, rates it by what it touches and asks you first. On by default; only you can turn it off.
Measured, not promised
A 3-line bug went from 335 s · USD 0.339 to 77 s · USD 0.105 once small work stopped paying for delegation.
The stack
Everything an agent needs to behave like a senior engineer.
Five open-source tools. Most teams start with gentle-ai and add the rest when they need them.
gentle-aiStart here
The ecosystem configurator
Equips the agents already on each machine with memory, a workflow, skills, MCP servers, a persona and review guardrails. Never installs an agent for you.
engram
Persistent memory for every agent
One local Go binary with SQLite and full-text search. Git Sync shares memory across machines; Engram Cloud adds optional, self-hosted, project-scoped replication with a dashboard.
gentle-shell
A Pi-native workspace for AI-assisted development
Turns a Pi session into a workspace you lead: changes, agents, commands and permission modes, with Gentle AI inside and a double-click installer.
gga
AI code review before the commit leaves the machine
A pre-commit review with the provider you already have: Claude, Gemini, Codex, OpenCode or Ollama.
Gentleman.Dots
The same environment on every desk
One TUI installer for Neovim, your shell, multiplexer, terminal and the Starship prompt, on macOS, Linux, WSL and Termux.
Install
From a fresh laptop to a disciplined agent in three steps.
brew install gentleman-programming/tap/gentle-ai# one formula on macOS and Linuxgentle-ai# pick your agents, a preset and a persona in the TUIgentle-ai doctor# read-only health report: binaries, state, memory, disk
- Detects what you already use. It configures the agents on the machine and never installs one for you.
- Backed up first. Every install, sync and upgrade snapshots your configuration before writing.
- Windows is fully tested and builds from source with Go, never from an unsigned download.
- Stay current with
gentle-ai upgradeandgentle-ai sync.
FAQ
The questions engineering leads ask first.
Still unsure? Write to me and I reply personally.
How to contact me →Gentle AI, Gentle Shell, gentle-pi: what is the difference?
Gentle AI configures the coding agents you already use (Claude Code, OpenCode, Cursor, Codex, Pi and more) with memory, the ODD workflow, skills and review guardrails, and it never installs an agent for you. Gentle Shell is a complete workspace built on Pi, with its own Pi runtime, agents and interface (Gentle Changes, Gentle Agents, Gentle Commands and the permission modes); its installer bundles its exact Pi, a pinned Gentle AI and Engram. gentle-pi is not a third product: it is Gentle Shell's npm package and repository name until the migration. In short: Gentle AI improves the agent you already have; Gentle Shell is the full workspace on Pi, with Gentle AI inside.
Does Gentle-AI install AI agents for us?
No. It configures agent runtimes already on the machine. If you select one it cannot detect, it refuses and prints the exact command you would run yourself. It never installs software silently.
Which agents does it support?
Full delegation, where focused subagents do the work: Claude Code, OpenCode, Kilo Code, Gemini CLI, Cursor, VS Code Copilot, Kimi Code, Kiro IDE and Qwen Code, plus Pi through Gentle Shell. Solo-agent, everything in one conversation with the same persona, skills, memory and workflow: Codex, Windsurf, Antigravity, OpenClaw and Trae.
Where does our memory live, and who can see it?
Locally, in a SQLite database under ~/.engram that stays authoritative. Git Sync exports chunks you control, and Engram Cloud is optional, self-hosted replication you enrol per project. Gentle-AI does not send your code anywhere.
Will it slow down small tasks with process?
No. Understood, low-risk work runs inline, with no subagents and no task document. In the benchmark a 3-line bug went from 335 s to 77 s and from USD 0.339 to USD 0.105.
Is the evidence-based review step mandatory?
No. It is on by default and you turn it off with gentle-ai review mode disable. A review is only due for high-risk changes, and its outcome is informational: it never blocks a commit or a release.
Why is it so opinionated?
Most agent sessions fail for operational reasons, not model reasons. It ships a working set of opinions (requirements first, tests as evidence, reviewable work units, security defaults) so the team stops re-litigating them; the Custom preset keeps your own persona and settings.
How do we roll it out to a team?
Everyone installs the same binary and picks the same preset, so every machine gets the same baseline. Start with a two-week pilot on one team and one measurable question: how much rework, repeated context and review time did it remove?
Is it open source, and does it run on Windows?
Yes. gentle-ai, engram, Gentle Shell and Gentleman Guardian Angel are MIT; Gentleman.Dots is Apache-2.0. Gentle-AI is fully tested on Windows and builds from source with Go 1.25.10 or newer; Gentleman.Dots targets Windows through WSL.
How do we update, and what if something breaks?
Run gentle-ai upgrade followed by gentle-ai sync. Every change snapshots your configuration first, keeping the five most recent backups, and gentle-ai doctor is a read-only report you can run any time.
Who is behind this, and how to work with him
Built by someone who spent 15 years in enterprise architecture, and then built the tools.
- 15+
- years of enterprise architecture
- 5
- open-source tools, built and used daily
- 1000s
- of developers in the community
Alan BuscagliaGentleman Programming
The Gentle stack is not a consultant's slide deck. It is the toolset Alan Buscaglia uses every day, refined in public with a community of thousands of developers who test it on real code.
Teams usually reach him the same way: a CTO or tech lead discovers the channel, wants their developers to adopt AI, and it is not working. Reluctant devs, disorderly usage, everyone prompting their own way, no process and no quality bar. That is an adoption problem, not a prompting course.
The focus is quality and verification over raw speed, and the goal is a team that knows how to direct the AI, not one that depends on a consultant.
- 15 years of enterprise architecture before writing a single line of these tools.
- Open-source tools he built and uses himself: gentle-ai, Engram, Gentle Shell, the 4R review framework.
- A community of thousands of developers testing everything on real projects.
- Verification over generation: quality first, speed as a consequence.
- The team learns to lead the AI; the engagement is designed to end.
Ways of working together
Diagnosis, four sessions and a deliverable, spread over the rollout
AI adoption program
For teams that want to adopt AI with a process behind it. Combines group sessions with tech-lead alignment before and after the rollout.
- Initial diagnosis of the team: technical level, resistances, opportunities.
- Group session before the rollout: AI as a tool, the Tony Stark and Jarvis approach, demystification, real cases.
- Tech-lead session before the rollout: aligning workflows and the adoption strategy.
- Tech-lead session after the rollout: follow-up and friction points.
- Group session after the rollout: consolidating what was learned and next steps.
- Deliverable: documentation and workflows tailored to the team's stack.
Three sessions on your own code
Hands-on with your codebase
No theory. The team sees the whole flow, gentle-ai, Engram and the 4R review, running on the problems they actually have.
- Survey of real problems with the developers, and an assignment for them.
- Live demonstration with gentle-ai and the agentic ecosystem solving exactly those problems.
- The developers show how they are using it and get calibrated recommendations.
Longer engagements, delivered phase by phase
Phased engagement
For organisations that need adoption to land across several teams or quarters, with a clear scope and outcome per phase.
- Scope and success criteria per phase.
- Delivery with the team's own tools and code.
- Review and decision before the next phase.
Speaking
On stage, on the topics this stack is built around.
Keynotes and technical talks at the main developer conferences in Spain.
The 94,000-line pull request, and how Receipt-Driven Development was born
AI Afterwork: Agentic Stack · InnoIT Consulting offices, Barcelona
- Keynote
Keynote, invited by Talent Arena and Mobile World Capital Barcelona
Talent Arena Night · Antiga Fàbrica Estrella Damm, Barcelona
Code review in the age of vibe coding
Codemotion Madrid 2026 · MEEU Chamartín, Madrid
4 earlier talks
- Workshop
From vibe coding to software engineering with AI: agent architecture for scalable projects
JSConf España 2026 · La Nave, Madrid
AI agents as co-maintainers: creating and maintaining open-source projects with AI
GDG Barcelona: Build with AI · eDreams ODIGEO, Barcelona
Beyond the framework: why code matters less in the age of AI
DevFest Barcelona 2025 · 42 Barcelona, GDG Barcelona
Code review in the age of vibe coding: how seniors become quality firewalls
Nerdearla Madrid · La Nave, Madrid
Topics · Formats
- Code review in the age of AI
- Agent architecture and orchestration
- From vibe coding to engineering
- Concepts over code
- Keynote · Talk 30–45 min · Workshop
Give your agents a brain and a process.
One command per machine. Open source. Nothing to unlearn if you stop using it.