Scoped API Key Pattern

Scoped API Key Pattern

The security practice of creating a dedicated account or API key per AI agent, with the minimum permissions that agent actually needs. Prevents an agent from having unintended write access, enables per-agent spend tracking, and isolates blast radius if credentials are leaked.

Key points

  • Create a separate account (e.g., UpAI in ClickUp) rather than sharing your personal credentials with Claude Code [src-013]
  • Set API permissions to the minimum required: if the AIOS only needs to read tasks and create comments, don't grant full admin access [src-013]
  • Per-account API keys enable spend tracking by agent — useful for billing analysis and debugging runaway automation [src-013]
  • Nate: "I created an account called UpAI. And now I give my UpAI API key to Claude Code rather than my own personal API key. Per API key or per account, you can set different permissions." [src-013]
  • In client delivery, Nate extends the pattern to ownership: the client should own API accounts and billing from the start, while the builder guides setup and avoids becoming the billing intermediary [src-016]
  • Secrets belong in environment/config layers such as .env, not embedded in CLAUDE.md, workflow text, or handover documentation [src-016]
  • For Hermes, Nate recommends treating each assistant like a new employee or intern: create named accounts and keys for the agent, give only the permissions it needs, and keep OpenRouter, Perplexity, Telegram, GitHub, and other credentials scoped and observable [src-074].
  • This becomes more important as agents multiply. Separate Hermes containers should not all reuse the same broad personal API keys because spend, mistakes, and compromise become harder to attribute [src-074].
  • Roberts's Apollo skill example extends the same pattern to API-backed sales-data connectors: store credentials in environment variables, expose only the workflow the agent needs, and avoid putting secrets into memory or prompts [src-079].
  • For Gmail and Calendar connectors, Roberts explicitly favors least access: draft email before send authority, and calendar permissions chosen around the actual assistant job [src-079].

Related entities

  • ClickUp — primary example: UpAI account with scoped ClickUp API key
  • Anthropic — recommends principle of least privilege for agent credentials

Related concepts

Source references

  • [src-013] Nate Herk — "Build & Sell Claude Code Operating Systems (2+ Hour Course)" (2026-05-01)
  • [src-016] Nate Herk — "Build & Sell with Claude Code (10+ Hour Course)" (2026-03-12)
  • [src-074] Nate Herk — "Hermes Agent: Zero to Personal AI Assistant (1 Hour Course)" (2026-05-10)
  • [src-079] Jack Roberts — "Hermes Agent just got 10X Better (Agentic OS)" (2026-05-15)

Robin Cartier perspective

This page is part of Robin Cartier's working AI knowledge graph: a practical research layer for production AI, recommendation systems, experimentation, GEO, and agentic web readiness.

The useful next step is to connect this concept back to applied product leadership and operating models.

Recommended next

Keep reading from this thread

From 494 indexed pages and articles.

  1. Wiki concept ClickUp Key facts Type: Project management / task tracker API: REST API; Nate uses a dedicated UpAI ClickUp account with a scoped API key Role in AIOS Related by account
  2. Wiki concept Personal Agent Container Isolation The practice of scaling personal AI agents by giving each serious agent its own container, credentials, memory Related by permissions
  3. Insight AI Measurement and Experimentation How to measure AI product impact with evals, adoption metrics, online experiments, guardrails, and cost tracking Related by tracking