Glossary

One place for every term the docs use. Terms link to the page that treats them in depth.

Architecture

Webhook-sovereign — Token Vault's only architecture: credentials live on a webhook server you deploy and control; Token Vault brokers access but never holds, sees, or can decrypt a credential. See Architecture.

Control plane — the Token Vault service (api.tokenvault.uk): identities, grants, ABAC policies, audit metadata, and your webhook's URL + HMAC secret. Holds no credentials.

Data plane — your webhook: credential storage, encryption (your choice of), OAuth refresh execution, and upstream API calls. The sole credential custodian.

Webhook — a small HTTP server implementing the webhook protocol (/v1/… endpoints). Deploy the TypeScript reference or write your own from the machine-readable spec.

Signed ticket — a short-lived HMAC-SHA256-signed authorization Token Vault attaches when routing a request to your webhook (or redirecting an agent to it). The ticket proves Token Vault authorized the operation; the credential itself never passes through Token Vault.

Kill switch — take your webhook offline and every credential path stops instantly; there is nothing in Token Vault to fall back to. See Architecture → kill switch.

Bind / binding — the one-time handshake pairing a webhook with your account via a one-time code and /v1/exchange; the HMAC secret is established automatically. See Binding.

Capability — a feature flag your webhook advertises during binding (e.g. tv-refresh, totp) that unlocks optional flows.

Identities & keys

tvagent_… — an agent API key. Identifies one agent; passed as Authorization: Bearer, x-agent-key, or (deprecated) ?key=. See Agents & Grants.

mcp_… — an MCP proxy key. Identifies one proxy configuration; revocable in one click. See MCP Proxy.

tvsess_… — a one-hour OAuth 2.1 session token bound to one (user, agent) pair, obtained in-client via the MCP OAuth flow. Same grants and policies as the agent's static key.

tvrefresh_… — the refresh token paired with a tvsess_…; every refresh re-checks the agent's status, so suspending an agent kills its live sessions within the hour.

Agent — an identity you create for each AI agent, script, or service that needs credentials. Suspend or delete it to cut access instantly.

Access control

Grant — permission for one agent to access one stored credential, with an expiry (1 hour to "until revoked"). Expired grants are filtered and deleted automatically.

Token group — a bundle of tokens grantable as a unit; grants created from a group carry its sourceGroupId.

ABAC policy — a reusable set of attribute-based rules (time window, IP allowlist, rate limit, usage cap, geo-restriction, manual approval) attached to agents, proxies, or tokens. Rules within a policy AND together; policies on an entity AND together. See Policies.

Manual approval — a policy rule that pushes an approve/deny prompt to your phone before a credential is released.

Tokens

Token / credential — the docs use these interchangeably for a stored secret: an OAuth access/refresh token pair, an API key or PAT, a TOTP seed, or a raw credential (e.g. a GCP service-account JSON).

TOTP token — a stored TOTP seed; agents receive the current 6-digit code, never the seed. See TOTP guide.

Raw credential brokering — storing a long-lived credential (GCP SA JSON, AWS keys) from which your webhook mints short-lived access tokens for agents. See the guide.

TV Zero / TV Refresh — the dashboard trust badges: green for tokens Token Vault never touches in any flow; blue for tokens using the opt-in TV-mediated refresh (in transit only, never stored).