# Golden Path

> Editorial library on designing environments where AI agents do good work without taking production down: concepts, patterns, guides and anonymized real-world cases.

## Agent access

- [MCP server](https://ia.dmaledicte.cloud/mcp): Streamable HTTP, public and read-only. Serves English by default. Setup: https://ia.dmaledicte.cloud/en/conectar
- [Full library as JSON](https://ia.dmaledicte.cloud/en/content.json): every entry with its original Markdown
- [Glossary](https://ia.dmaledicte.cloud/en/glosario): reference vocabulary
- [Versión en español](https://ia.dmaledicte.cloud/llms.txt): this index in the original Spanish

## Agent governance

- [Guided models — the environment is the policy](https://ia.dmaledicte.cloud/en/entrada/modelos-guiados-el-entorno-es-la-politica): Working safely with an agent depends less on the model remembering to ask for permission and more on the harness, the operating system and the credentials making it impossible to act outside the intended scope.
- [CLI vs. web — the agent inside the project](https://ia.dmaledicte.cloud/en/entrada/terminal-vs-web-superficies-trabajo): The terminal does not make the model smarter: it places it inside the system it has to understand, modify and verify. The web keeps the edge for thinking, researching and communicating.
- [Choosing a model and reasoning level](https://ia.dmaledicte.cloud/en/entrada/guia-escalado-modelos): A practical guide to assigning coordinators and workers, raising reasoning before paying for a bigger model and scaling up when a real capability gap appears.
- [Before publishing, test the real startup](https://ia.dmaledicte.cloud/en/entrada/preflight-despliegue-backend): A successful build does not guarantee the backend can start. A preflight with the production command and a blocking healthcheck close that gap.
- [Amp and the cloud agent factory](https://ia.dmaledicte.cloud/en/entrada/amp-fabrica-agentes-cloud): Amp packages agent, machine, context and Git into a single surface. It does not invent multi-agent development, but it turns into a product a practice we usually still build by hand.
- [How to read the OpenRouter rankings](https://ia.dmaledicte.cloud/en/entrada/como-leer-rankings-openrouter): A guide to interpreting OpenRouter usage, spend, speed and benchmarks, following its thirteen text sections and turning the signals into comparable tests.
- [Jev: small decisions inside large systems](https://ia.dmaledicte.cloud/en/entrada/jev-decisiones-estructuradas): What a structured-decision model does and where it belongs: routing, classification, scoring and gates before state changes.
- [Gentle AI and Engram: memory and process for coding agents](https://ia.dmaledicte.cloud/en/entrada/gentle-ai-engram-memoria-y-proceso): What an environment configurator contributes and what persistent memory solves when agent work spans sessions, tools and projects.
- [Tailscale: a private network built around identity](https://ia.dmaledicte.cloud/en/entrada/tailscale-red-privada-identidad): Tailnets, WireGuard, direct connections, and relays: what Tailscale solves and what remains your responsibility.
- [Tailscale: permissions you can explain](https://ia.dmaledicte.cloud/en/entrada/tailscale-permisos-grants): Design grants by identity, destination, and port; test allowed and denied access; separate networking, SSH, and application permissions.
- [Comparing CLI harnesses: Codex, Claude Code, OpenCode V2, Cline and OMP](https://ia.dmaledicte.cloud/en/entrada/comparar-harness-cli-codex-claude-opencode-cline): The strengths, differences and drawbacks of five coding harnesses: how to choose them, assign responsibilities and assess the real cost of agent work.
- [Advanced agent routing: model, effort, role and harness](https://ia.dmaledicte.cloud/en/entrada/enrutamiento-modelo-esfuerzo-rol-harness): Why agent selection must account for model, reasoning effort, task role and execution harness, using Codex versus OMP as a controlled contrast.
- [RCT and Overdrive: human protocols for managing provider quotas](https://ia.dmaledicte.cloud/en/entrada/protocolos-humanos-cuota-rct-overdrive): Two prepared responses to scarce or expiring capacity: protect LIVE operations, prioritize useful BUILD work, and keep control with the human operator.

## Orca

- [Orca as an ADE and control plane](https://ia.dmaledicte.cloud/en/entrada/orca-como-ade): Orca organizes agents, repositories and execution environments; it does not replace the model or decide on its own how the team should work.
- [Workspaces, projects and worktrees](https://ia.dmaledicte.cloud/en/entrada/workspaces-y-worktrees): The project organizes the repository, the workspace represents a work front and the worktree provides real isolation of files and branch.
- [Orchestration in Orca](https://ia.dmaledicte.cloud/en/entrada/orquestacion): Orchestrating is not opening agents in parallel: it is distributing authority, dependencies, context, decisions and evidence of completion.
- [SSH hosts and execution topology](https://ia.dmaledicte.cloud/en/entrada/hosts-ssh): Orca can act as a local interface while agents, repositories and runtimes live in WSL or on a remote machine.
- [Agent terminals and the session lifecycle](https://ia.dmaledicte.cloud/en/entrada/agent-terminals): An agent terminal ties together a CLI, a session, a worktree and an observable state; it is not just a console under another name.
- [Continuity, recovery and observability](https://ia.dmaledicte.cloud/en/entrada/continuidad-y-observabilidad): Orca preserves layout and scrollback, and can keep agents alive as long as the daemon survives; critical state must still exist outside the interface.
- [Orca Desktop with Windows and WSL](https://ia.dmaledicte.cloud/en/entrada/setup-windows-wsl): A reference pattern for separating the desktop interface from the Linux environment where repositories, runtimes and agents live.
- [Portable access to Orca over a private network](https://ia.dmaledicte.cloud/en/entrada/acceso-portatil-orca): Use another device as a client of the same development environment, without duplicating repositories or exposing SSH to the Internet.
- [Case study — Orca as a multi-agent cockpit](https://ia.dmaledicte.cloud/en/entrada/caso-orca-cockpit-multiagente): The evolution from one coordinator to separate direction, integration and packages, with cross-family auditing and verifiable acceptance.
- [Planning a scalable AM5 workstation](https://ia.dmaledicte.cloud/en/entrada/planificar-workstation-am5-escalable): How to audit an existing PC and migrate to AM5 across three stages — budget, mid-range and high-end — without rebuying half the machine at each step.

## Agentic design

- [OpenDesign and the visual direction of agents](https://ia.dmaledicte.cloud/en/entrada/opendesign-direccion-visual-agentes): Models can already build interfaces. The problem is getting them to stop improvising the visual identity on every run. OpenDesign proposes turning that direction into portable files.
- [OpenDesign across Windows, WSL, plans and proxies](https://ia.dmaledicte.cloud/en/entrada/opendesign-integracion-local-cli-proxy): Local-first does not always mean a single environment. When the interface lives on Windows and the agents in WSL, choosing between CLI, BYOK and proxy defines spend, sessions and risk surface.

## Hermes

- [Hermes as an operational agent](https://ia.dmaledicte.cloud/en/entrada/hermes-agente-operativo): A self-hosted agent that brings queries and alerts to the channels where the team works, without turning every interaction into a technical session.
- [Hermes installation preset](https://ia.dmaledicte.cloud/en/entrada/hermes-preset-instalacion): Five decisions to make before setting up Hermes without copying someone else’s infrastructure: hosting, model, channels, concurrency and integrations.
- [Case study — Hermes as an operational interface](https://ia.dmaledicte.cloud/en/entrada/caso-hermes-operaciones): A self-hosted agent brought queries and alerts into everyday channels, with scoped permissions and observable business tests.

## Versioned CRM

- [From backup to a CRM function platform](https://ia.dmaledicte.cloud/en/entrada/backup-versionado-crm): The read-only backup became the foundation of an internal platform for analyzing, reviewing and deploying CRM functions with separate authorities.
- [Read-only as an architectural decision](https://ia.dmaledicte.cloud/en/entrada/arquitectura-read-only): Keep backup and analysis without production authority while a separate path allows reviewed, verifiable changes to be deployed.
- [From backup to process map](https://ia.dmaledicte.cloud/en/entrada/mapa-procesos-verificacion): The process map enables analysis without production access; the new change path connects verifiable findings with review and controlled deployment.

## Product cases

- [Case study — From counter to operational continuity dashboard](https://ia.dmaledicte.cloud/en/entrada/caso-tablero-continuidad-operativa): A personal tool grew to combine incidents, service health and progressive signals without bringing credentials into the browser.
- [Case study — Widgets that turn CRM rules into guided flows](https://ia.dmaledicte.cloud/en/entrada/caso-widgets-crm-flujos-guiados): Two embedded widgets brought context, validation and automation to the exact point where the team made decisions.
- [Case: a Director that separates narration from memory](https://ia.dmaledicte.cloud/en/entrada/caso-jev-director-narrativo): An orchestrated architecture where one model writes the scene, Jev decides its consequences and an event log rebuilds what each character can remember.
- [Tailscale for an agent environment: laptop, VPS, and WSL](https://ia.dmaledicte.cloud/en/entrada/tailscale-entorno-agentes): An anonymized reference design for accessing a workstation, operating a VPS, and sharing internal previews without exposing admin ports.