Last updated 2026-04-16. These apply when an AI agent calls MeterCall on behalf of a human or organization (the "principal").
Core rule: the principal is responsible for the agent's actions. The agent is a tool. The human or org behind it is accountable.
Who is the principal?
The principal is the MeterCall account holder whose wallet, budget, or subscription funds the agent's calls. That can be a person, a startup, or an enterprise. When an agent authenticates via an agent-scoped key, it inherits the principal's identity for billing and compliance.
Principal responsibility
You pay for what your agent spends. Including overages above any cap you set, if your cap was soft or you raised it.
You own the agent's outputs. You're responsible if it defames, infringes, or harms.
You set guardrails. Spending caps, allowed catalogs, geographic scopes — these are your knobs. If you leave them wide open, you own the blast radius.
You answer disputes. If an upstream vendor disputes a call your agent made, you're the one who talks to them. MeterCall provides logs.
Spending caps
Every agent key must have a cap. Caps have two modes:
Hard cap: MeterCall refuses the call past the cap. No overage. Recommended for autonomous agents.
Soft cap: MeterCall alerts you and keeps serving. Principal pays any overage on the next invoice.
Caps reset on the interval you pick (per-call, daily, weekly, monthly). If you don't pick, default is daily hard cap at $10.
Agent identity must be declared
When an agent calls a human-facing surface through MeterCall — email, SMS, chat, social — it must identify as an agent in a way the recipient can detect. Specifically:
Do not impersonate a named human without that human's written consent held by the principal.
Outbound messages should include an agent marker (footer, handle suffix, or protocol-level signal) that the recipient's system can surface.
Voice calls originating from agents must comply with the receiving jurisdiction's AI-disclosure law (many US states, EU AI Act).
Rate limits for agents
Agents run hot. We give them their own ceiling on top of the principal's tier:
Burst: 300 requests/min per agent key, regardless of tier, unless the principal raises it explicitly.
Parallel caps: max 50 concurrent in-flight requests per agent key by default.
Anomaly circuit breaker: if spend accelerates 10x over the rolling 24h average, we pause and alert the principal.
Dispute resolution
Disputes involving agent behavior are principal-led. If a third party complains about your agent, we route them to you. If you complain about an upstream's handling of your agent's traffic, we help you escalate but you are the party of record. Arbitration clause from the main Terms applies.
Logging + audit
Agent-key traffic is logged at the metadata level (catalog, timestamp, status, spend) for 13 months for audit. Bodies remain off by default — see Privacy. You can export your agent's audit log any time from the dashboard.
Revocation
You can revoke an agent key instantly from the dashboard or POST /api/agents/{id}/revoke. Revocation is effective within 5 seconds globally. In-flight calls complete.