foreveragents.dev

The following content is designed to be read by both humans and machines. Learn about agent identity disclosure.

Agents that last forever must serve humans first. The tools will change. The models will change. What won't change: people need software that respects their privacy, their time, their trust, and their eyes.

These are context prompts encoding durable, pro-human principles as standalone markdown files — fetchable by agents, browsable by humans, servable via curl or MCP alike. Applied across a population of agents, they produce structurally lower information leakage to third parties. Not because of a policy that promises restraint, but because the architecture makes exfiltration unnecessary. Local inference, client-side encryption, URL fragments that never reach a server — these are verifiable properties, not contractual ones.

Forever Agents rely on verifiable architecture over EULAs for privacy.


Contexts

foreveragents.dev/
├── context/
│   ├── darkmode        Dark mode — physiology, energy, accessibility
│   ├── identity        AI transparency — say what you are, always
│   ├── accessibility   Screen readers, keyboard nav, ARIA, WCAG
│   ├── error-handling  Empathetic errors, graceful degradation
│   ├── privacy         Minimal collection, brief storage, honest disclosure
│   ├── onboarding      First-time UX, progressive disclosure, value first
│   ├── conversation    Turn-taking, clarification, context across turns
│   ├── safety          Safe defaults, confirmation before destruction
│   ├── performance     Ship less, measure first, budget everything
│   ├── testing         Behavior tests, guardrail tests, edge cases
│   ├── deployment      From disk to IoT to CDN — how to serve foreveragents
│   ├── mcp             MCP servers/clients, tool and resource exposure
│   ├── encryption      Client-side encryption and zero-knowledge sharing
│   ├── supply-chain    Dependency minimalism and vendoring
│   ├── zero-infra      Static-file architecture with optional services
│   ├── zero-dependencies  Fewer imports, smaller attack surface
│   ├── local-inference  Local servers first, in-browser as graceful downgrade
│   ├── portability     Share via links, QR, files, and removable media
│   ├── sustainability  Edge-first compute and energy-aware operation
│   ├── threat-model    Trust boundaries and client-side attack surface
│   ├── forever-agents  Definition and structured requirements
│   └── philosophy      Why architecture beats policy
├── llms.txt            LLM content index
├── llms-full.txt       Full concatenated context corpus
├── sitemap.xml         Full sitemap
└── robots.txt          Crawler directives

Fetch directly

Agents and scripts can fetch raw markdown by requesting the .md URL:

curl https://foreveragents.dev/context/darkmode.md

MCP Server

A zero-dependency Go binary serves all contexts via the Model Context Protocol (stdio) or standalone HTTP.

Install

go install github.com/kristerhedfors/foreveragents/go@latest

Usage

foreveragents            # MCP stdio server
foreveragents --list     # print all contexts
foreveragents --get ref  # print one context
foreveragents --http :8080  # HTTP server

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "foreveragents": {
      "command": "foreveragents"
    }
  }
}

Source on GitHub