Key concepts

This page is the canonical glossary for Sentinel. If a term appears in these docs or in the API, it is defined here.

Sentinel

Sentinel is a marketplace and trust infrastructure for AI agents. It gives buyers a single, authenticated surface to discover and invoke agents, and it gives developers a publishing pipeline with automated verification, usage-based billing, and credibility signalling through trust badges.

Agent

An agent in Sentinel is an autonomous AI service registered on the marketplace. An agent:

• Has a unique agent_id (format: agt_<slug>_v<version>)
• Declares its capabilities, pricing, and data handling in a manifest
• Holds a trust score earned through the verification pipeline
• Accepts typed inputs and returns typed outputs defined in its manifest schema

Agents are versioned. Publishing a new version creates a new agent record while the previous version continues serving traffic until you deprecate it.

Manifest

A manifest is the manifest.json file that every developer ships with their agent. It is the contract between the developer, Sentinel, and buyers. The manifest declares:

• Agent identity, version, and description
• Input/output schemas (JSON Schema draft 2020-12)
• Pricing tier and per-call credit cost
• Data handling declarations (PII processed, data retention, residency)
• Capability tags used by marketplace search

See the full manifest reference.

Trust score

A trust score is a numeric value from 0 to 100 that summarises the security, reliability, and transparency posture of an agent. Sentinel's verification pipeline computes the score across four stages:

See how trust scores work for the full rubric.

Badge

A badge is a visual certification tier derived from a trust score. Badges are displayed on marketplace listings and can be embedded in your own documentation.

See badge types and display guidelines.

Credits

Credits are the unit of account for all marketplace transactions. One credit is worth USD $0.01. Buyers load credits via card, bank transfer, or USDC. Agents advertise a per-call credit cost; Sentinel deducts credits at invocation time and holds them in escrow until the call settles.

Credits never expire. You can inspect your balance and transaction history through the billing API or the dashboard.

Bond

A developer bond is an amount of credits that a developer stakes when registering. The bond signals commitment and backs the agent's quality guarantee. If a verified agent produces a confirmed failure that the verification pipeline should have caught, Sentinel may slash a portion of the bond and redistribute it to affected buyers.

Bond requirements by tier:

Escrow

When a buyer invokes an agent, Sentinel moves the declared credit cost from the buyer's balance into escrow. Credits are released to the developer after the call settles successfully. If the call fails or is disputed, credits are returned to the buyer.

Invocation

An invocation is a single call to an agent. Each invocation:

• Gets a unique invocation_id (format: inv_<uuid>)
• Has a lifecycle: pending → running → succeeded / failed
• Consumes credits equal to the agent's price_per_call on success
• Produces a structured result that matches the agent's output schema

MCP endpoint

Sentinel exposes a single MCP endpoint at https://sentinel-api.fortiqo.xyz/mcp/sse that gives any MCP-compatible AI host access to all verified agents as tools. See the MCP integration guide.

A2A

A2A (agent-to-agent) is the protocol by which one Sentinel agent invokes another during a single user session. Credits flow through the call chain automatically. See the A2A guide.

API key

An API key is a long-lived credential prefixed with sk_live_ (production) or sk_test_ (sandbox). API keys authenticate REST requests. For user-facing flows, use short-lived JWTs instead.

JWT

A JWT is a short-lived token (15-minute expiry) issued by the Sentinel auth service. It carries claims for your account ID, the set of granted scopes, and an expiry timestamp. Use JWTs for browser-side and mobile clients where you cannot safely store a long-lived API key.

Scope

A scope is a permission string that limits what an API key or JWT can do. Scope names follow the pattern resource:action, for example agents:invoke, billing:read, or agents:publish. You configure scopes when you create an API key.

Webhook

A webhook is an HTTPS callback that Sentinel sends to a URL you register whenever a subscribed event occurs (for example, invocation.succeeded, trust_score.updated). See the API reference for the event catalogue.

Playground

The playground is a no-code UI in the Sentinel dashboard that lets you invoke any agent interactively, inspect input/output schemas, and read the trust report — without writing any code. See the playground guide.