Operate · Policy
Guardrails designer
Set budgets and rate limits at three scopes. Export a policy your AI gateway can enforce.
Per engineer
The most important scope. Catches runaway scripts and over-eager agents.
Soft warn threshold80% of cap
Per tool
Cap each integration independently. Catches one tool eating the whole budget.
GitHub Copilot Business
GitHub
$/seat/mo
Cursor Pro
Anysphere
$/seat/mo
Claude Code (Max)
Anthropic
$/seat/mo
ChatGPT Business
OpenAI
$/seat/mo
Anthropic API (raw)
Anthropic
$/seat/mo
OpenAI API (raw)
OpenAI
$/seat/mo
Model allowlist
Premium models often cost 10–60× cheaper alternatives. Allowlist explicitly.
4 of 10 models allowed. Disallowed models will be rejected at the gateway with a clear error.
Per API key & MCP server
Stop a single key or MCP from torching the bill.
Tip: For MCP servers, also enforce a max tool-result size (≤ 4KB recommended) and a max agent step count (≤ 30) — both common runaway-cost vectors.
Target gateway
Tokmeter is the default — one click to apply. Other targets are for teams migrating off an existing gateway.
★ Default. Tokmeter applies these guardrails natively across every connected provider — no proxy required.
Policy preview
tokmeter compatible YAML
Default target — no deploy needed. These rules apply natively across every connected provider. Export to LiteLLM, Portkey, or Helicone only if you are migrating off an existing gateway and need the same policy in their format.
# Tokmeter native guardrails policy
# Apply with one click — no proxy or YAML deploy required.
# This file is for audit/source-control; Tokmeter persists the same rules in your workspace.
policy_version: 1
workspace: default
per_engineer:
daily_usd: 50
monthly_usd: 600
soft_warn_pct: 80
enforcement: soft # soft = warn + notify; hard = block at cap
per_tool_seat_caps_usd:
cursor: 40
claude-code: 200
github-copilot: 19
chatgpt-business: 25
anthropic-api: 300
openai-api: 250
model_allowlist:
- claude-sonnet-4
- claude-haiku-4
- gpt-5-mini
- gemini-2.5-flash
request_limits:
rpm: 60
max_tokens: 50000
notifications:
channels: [in_app, email, slack] # configure under Settings → Integrations
Estimated impact
Max possible monthly spend / engineer$600.00
Annual cap @ 150 engineers$1.1M
Hard block enforcementOff (warn only)