ADE
Agent Development Environment: an environment built to run, organize and supervise development agents.
Full library
Full index
All 35 Golden Path entries, grouped by path. Each one has its own page and can be copied or downloaded as Markdown.
← Back to the catalog⚿
Agent governance
13- ConceptGuided models — the environment is the policyWorking 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.
- ConceptCLI vs. web — the agent inside the projectThe 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.
- GuideChoosing a model and reasoning levelA practical guide to assigning coordinators and workers, raising reasoning before paying for a bigger model and scaling up when a real capability gap appears.
- GuideBefore publishing, test the real startupA successful build does not guarantee the backend can start. A preflight with the production command and a blocking healthcheck close that gap.
- EditorialAmp and the cloud agent factoryAmp 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.
- GuideHow to read the OpenRouter rankingsA guide to interpreting OpenRouter usage, spend, speed and benchmarks, following its thirteen text sections and turning the signals into comparable tests.
- ConceptJev: small decisions inside large systemsWhat a structured-decision model does and where it belongs: routing, classification, scoring and gates before state changes.
- EditorialGentle AI and Engram: memory and process for coding agentsWhat an environment configurator contributes and what persistent memory solves when agent work spans sessions, tools and projects.
- ConceptTailscale: a private network built around identityTailnets, WireGuard, direct connections, and relays: what Tailscale solves and what remains your responsibility.
- GuideTailscale: permissions you can explainDesign grants by identity, destination, and port; test allowed and denied access; separate networking, SSH, and application permissions.
- EditorialComparing CLI harnesses: Codex, Claude Code, OpenCode V2, Cline and OMPThe strengths, differences and drawbacks of five coding harnesses: how to choose them, assign responsibilities and assess the real cost of agent work.
- EditorialAdvanced agent routing: model, effort, role and harnessWhy agent selection must account for model, reasoning effort, task role and execution harness, using Codex versus OMP as a controlled contrast.
- EditorialRCT and Overdrive: human protocols for managing provider quotasTwo prepared responses to scarce or expiring capacity: protect LIVE operations, prioritize useful BUILD work, and keep control with the human operator.
◉
Orca
10- ConceptOrca as an ADE and control planeOrca organizes agents, repositories and execution environments; it does not replace the model or decide on its own how the team should work.
- ConceptWorkspaces, projects and worktreesThe project organizes the repository, the workspace represents a work front and the worktree provides real isolation of files and branch.
- ConceptOrchestration in OrcaOrchestrating is not opening agents in parallel: it is distributing authority, dependencies, context, decisions and evidence of completion.
- ConceptSSH hosts and execution topologyOrca can act as a local interface while agents, repositories and runtimes live in WSL or on a remote machine.
- ConceptAgent terminals and the session lifecycleAn agent terminal ties together a CLI, a session, a worktree and an observable state; it is not just a console under another name.
- ConceptContinuity, recovery and observabilityOrca preserves layout and scrollback, and can keep agents alive as long as the daemon survives; critical state must still exist outside the interface.
- GuideOrca Desktop with Windows and WSLA reference pattern for separating the desktop interface from the Linux environment where repositories, runtimes and agents live.
- GuidePortable access to Orca over a private networkUse another device as a client of the same development environment, without duplicating repositories or exposing SSH to the Internet.
- Case studyCase study — Orca as a multi-agent cockpitThe evolution from one coordinator to separate direction, integration and packages, with cross-family auditing and verifiable acceptance.
- GuidePlanning a scalable AM5 workstationHow 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
2- EditorialOpenDesign and the visual direction of agentsModels 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.
- EditorialOpenDesign across Windows, WSL, plans and proxiesLocal-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
3- ConceptHermes as an operational agentA self-hosted agent that brings queries and alerts to the channels where the team works, without turning every interaction into a technical session.
- GuideHermes installation presetFive decisions to make before setting up Hermes without copying someone else’s infrastructure: hosting, model, channels, concurrency and integrations.
- Case studyCase study — Hermes as an operational interfaceA self-hosted agent brought queries and alerts into everyday channels, with scoped permissions and observable business tests.
▣
Versioned CRM
3- ConceptFrom backup to a CRM function platformThe read-only backup became the foundation of an internal platform for analyzing, reviewing and deploying CRM functions with separate authorities.
- GuideRead-only as an architectural decisionKeep backup and analysis without production authority while a separate path allows reviewed, verifiable changes to be deployed.
- Case studyFrom backup to process mapThe process map enables analysis without production access; the new change path connects verifiable findings with review and controlled deployment.
◆
Product cases
4- Case studyCase study — From counter to operational continuity dashboardA personal tool grew to combine incidents, service health and progressive signals without bringing credentials into the browser.
- Case studyCase study — Widgets that turn CRM rules into guided flowsTwo embedded widgets brought context, validation and automation to the exact point where the team made decisions.
- Case studyCase: a Director that separates narration from memoryAn orchestrated architecture where one model writes the scene, Jev decides its consequences and an event log rebuilds what each character can remember.
- GuideTailscale for an agent environment: laptop, VPS, and WSLAn anonymized reference design for accessing a workstation, operating a VPS, and sharing internal previews without exposing admin ports.
Looking for a definition? Go to the glossary ·