← Blog
Guide · 2026-07-31 · 7 min

Team AI policy for multi-model chat (surface first)

Good policy does not ban models. It picks a default workspace, then rules for data, upgrades, and review.

A team AI policy for multi-model chat should start with the surface, not a permanent lab logo. On zy.ai, Free/Pro/Ultra share one chat product: free models first, frontier when needed, Ultra for Voice and research agents.

Policy pillars that actually stick

  1. Surface — default workspace (zy.ai) for history and handoffs
  2. Data — what may be pasted; secrets stay in vaults/MCP least privilege
  3. Models — free default; Pro for frontier; Ultra for agents/Voice
  4. Review — human owns shipping code, legal, and customer claims
  5. Training — short internal guide beats a 40-page PDF nobody opens

Why surface > single lab: single lab vs multi-model and multi-model access.

Recommended seat ladder

  • Most ICs: Free until they hit quality or limit walls
  • Heavy builders/writers: Pro ($15/mo)
  • Research/GTM leads: Ultra ($45/mo) for agents + Deep Research
  • Org billing: Corporate via contact@zyai.org

Upgrade signals: Free vs Pro · when free is enough.

Tools and agents under policy

  • MCP: allowlist trusted servers only — MCP tools
  • Business agents / Deep Research: Ultra, human edit required — business agents
  • Voice: Ultra; no sensitive dictation in shared spaces without process
Policy fails when it bans curiosity. It works when defaults are clear and exceptions are rare.

Rollout checklist

  1. Pick zy.ai as default chat URL in the wiki.
  2. Publish Free/Pro/Ultra decision table.
  3. Forbid pasting secrets; document MCP allowlist.
  4. Require human review for customer-facing and prod code.
  5. Revisit quarterly as model availability changes.

One-subscription framing for finance: one subscription many AI models. Docs: /docs.

Give the team one default surface

Start free on zy.ai. Add Pro/Ultra seats when weekly work earns them.