Capabilities

One policy and one configured scope per guard. All policy evaluation runs in process. With signing keys configured, decisions are Ed25519-signed and SHA-256 hash-chained over deterministic key-sorted JSON.

Capability	What you get
Cap windows	`per_call`, `per_minute`, `per_hour`, `per_day`, `per_month`. Any combination; each window has independent state. One policy can carry a per-minute burst guard, a daily soft cap and a monthly ceiling at once.
Cap actions	`allow` passes and logs. `shadow` passes and records the decision. `downgrade` rewrites the model parameter to a cheaper model, then proceeds. `block` raises `AgentGuardBlockedError` before the provider is contacted.
Scope dimensions	tenant, team, user, agent, task, provider. These form one collision-resistant scope key. Within one policy, the most restrictive action among exceeded caps wins.
Capability gating	`read_only`, `data_write`, `payment_initiate`, `payment_execute`. A policy can require a minimum tier; calls without a sufficient `capabilityClaim` are blocked before dispatch. Claims are caller-supplied unless the policy also requires a verified receipt-DAG attestation.
Localization	en-US, es-419, pt-BR. Block traces auto-detect the locale via `AGENTGUARD_LOCALE`, `LC_ALL` or the system locale. Override with `locale="es-419"`.
Verification	With `config.signingKeys`, each decision produces an Ed25519-signed entry. Anyone with the public key can verify without AgentGuard infrastructure. Without keys, enforcement still runs and the signed entry is `null`.
Audit	Entries are SHA-256 hash-chained. Tampering with any field invalidates the chain from that point forward. In-memory and NDJSON stores are exported; implement `DecisionLogStore` for other storage.
Provider portability	OpenAI, Anthropic and Bedrock native interfaces. The OpenAI binding also wraps compatible `chat.completions.create` clients, including OpenRouter-style endpoints.

Try it now

npm install -g @agentguard-run/spend
agentguard demo
agentguard verify

Core wrapper

import OpenAI from 'openai';
import { withSpendGuard } from '@agentguard-run/spend';

const client = withSpendGuard(new OpenAI(), {
  policy,
  scope: { tenantId: 'acme', agentId: 'agent-1' },
  config: { signingKeys, logStore },
});

CLI

Command	What it does
`agentguard demo`	Deterministic simulation that produces a real Ed25519 signature over real canonical JSON. No network calls.
`agentguard verify`	Verifies a signed receipt’s signature, entry hash and chain link.
`agentguard init`	Scaffolds `agentguard.policy.yaml` and `agentguard_quickstart.ts`, adds `.agentguard/` to `.gitignore`. Existing files are preserved unless you opt into overwrite.
`agentguard doctor`	Health check: crypto backend, provider SDK detection, keypair directory writable, policy file present, dry-run block simulation.
`agentguard explain <receipt>`	Pretty-prints a receipt: action, cap math step by step, what would unblock and when, signature breakdown.