Home

Appearance

Glossary ​

The words this documentation uses in one particular way, and the acronyms it assumes you know, each with the chapter that says the rest. A term here means the same thing on every page; when a page seems to use one differently, the page is wrong, not the glossary.

Terms ​

  • age — the file encryption format and tool the secret store is built on, and the kind of key rta kv init --generate makes. A name, not an acronym. Secrets
  • agent — the name an AI client is registered under, rta mcp serve --as claude. Grants are issued to that name and a lock freezes it, so consent given while talking to one client does not follow another. Also, loosely, the AI tool itself. Connect an agent
  • capability — one declared unit of work with an ID (kv.get, pg.query), typed inputs and a safety class. The CLI command, the TUI form and the MCP tool are all generated from that one declaration. Introduction
  • consent — answering a call that needs a grant nobody issued, while you are at the machine, instead of having it refused. A server started with --consent parks such a call and rta agent allow or rta agent deny answers it. Live consent
  • digest — the hash of a plugin's binary. Trust, and a grant's binding to a plugin build, are to a digest and never to a name or a version, so a rebuild asks again. Using plugins
  • forward — a kubectl port-forward or an ssh -L that rta opens for one call when a profile says a service is reached through a cluster or a host, and closes afterwards. Reaching private services
  • grant — permission for one capability, optionally one record, that expires on its own. A read that aims nowhere the agent chooses is free; everything that changes anything, and a read aimed at a destination the agent names, costs a grant. Grants
  • guard — the passphrase in front of issuing a grant, which makes "issue itself a grant" stop working for a process that only runs as you. Off by default. Grants
  • index — a git repository holding one manifest per plugin, which rta plugin index add points rta at and rta plugin install reads. The first-party one is official. Using plugins
  • instance — one of several connections a profile holds for the same plugin, named profile/instance (staging/analytics). The profile's default connection has no label. Profiles
  • lock — the instant no: freezes one agent, credential or operator now, ahead of any expiry or revocation, and lifts itself after a window if it was given one. Locks
  • operator — a person who manages a remote rta server by signing calls with a key only their passphrase can use, over the operator channel. The operator channel
  • plugin — a program that returns a declaration of capabilities and serves it over gRPC. Some are built into the binary; the rest are separate binaries from an index, each approved to run by its digest. Using plugins · Writing a plugin
  • policy — a team's file, .rta-policy.yaml, that no --ttl can argue with: a longest lifetime, targets never granted, connections never named, targets that need a named record. Team policy
  • profile — a named environment across every plugin that has something in it. staging points pg, s3 and vault at staging at once, and is also what a grant is scoped to with --profile. Profiles
  • record — two things, told apart by the article. The record is the chained, sealed log of every call that arrived over MCP and every authority change an operator made, read with rta agent log. A record is the one thing a call names — a key, an object, a host, a table — and what a grant can be narrowed to. The record · Grants
  • role — a list of grants written once, in a policy or in your config, that rta grant issue <role> issues to one agent under one passphrase prompt. Roles
  • root — a directory a path argument must sit under. The server's default is the directory it was started in, and --root widens it. The path gate
  • roster — a list rta keeps of who may do what: the grant roster that rta grant list shows, and on a remote server the operator roster, the file of enrolled operator keys. Grants · The operator channel
  • safety class — read, write or destructive, declared by each capability. It decides what an agent may reach without a person: reads by default, everything else with a grant. A capability that reveals a secret is write even though it changes nothing. Writing a plugin
  • store — the encrypted local secret store behind rta kv, locked to an age key or a passphrase. Secrets
  • surface — a way of reaching a capability: the CLI, the TUI or MCP. -o json, csv and markdown are formats the CLI answers in, not surfaces of their own. Introduction

Acronyms ​

  • CA — certificate authority; the issuer a TLS certificate chains up to.
  • CLI — command-line interface: rta typed at a shell.
  • CNPG — CloudNativePG, the PostgreSQL operator for Kubernetes that the cnpg plugin reads.
  • CWE — Common Weakness Enumeration, the numbered weaknesses (CWE-250) an rta audit finding cites beside its OWASP category.
  • DSN — data source name, a database's connection string; a profile spells the same thing as separate set: keys.
  • gRPC — the remote procedure call protocol a plugin serves its declaration over.
  • JSONC — JSON with comments, the format VS Code's mcp.json is written in, which is why rta prints its block instead of editing the file.
  • JWK — JSON Web Key, a key written as JSON; rta codec jwk reads one and says its type, size and thumbprint, and whether it is private.
  • MCP — Model Context Protocol, how an AI client discovers and calls tools. rta mcp serve is the server. MCP and the safety gate
  • OCI — Open Container Initiative; a plugin or the Helm chart can be published to an OCI registry and fetched from there. Using plugins
  • OIDC — OpenID Connect; an HTTP-hosted server can require a token from an identity provider. OIDC
  • OSV — the open vulnerability database rta audit deps asks.
  • OWASP — the Open Worldwide Application Security Project, whose Top 10 categories (A01:2025) an rta audit finding cites.
  • RBAC — role-based access control; in Kubernetes, what decides who may read which resource, and what rta's kube plugin leaves to the cluster. Kubernetes
  • RESP — the REdis Serialization Protocol; the redis plugin speaks it itself rather than linking a client library.
  • REST — the style of HTTP API that names a thing by its URL and acts on it with GET, POST, PUT and DELETE; rta http is a client for one.
  • RTA — short for Rule Them All, the name of the project; rta is the binary and how it is typed.
  • RWX — ReadWriteMany, a volume that several pods mount at once. The chart refuses it: two rta processes on one data directory would disagree about the grants. Kubernetes
  • SBOM — software bill of materials, the list of what a build contains; rta audit deps reads one, and a release publishes its own.
  • SIEM — security information and event management, the system a log is shipped to; the codes in the record are what to match on there, not the wording.
  • SLSA — Supply-chain Levels for Software Artifacts; the build provenance a release is published with. Verify a download
  • TLS — Transport Layer Security, the encryption beneath https and most database connections.
  • TTL — time to live; --ttl 30m is how long a grant, a profile or a lock stands before it lapses on its own.
  • TTY — a terminal. A form "at a TTY" is one a person fills in, as opposed to a script or a pipe, which has none.
  • TUI — terminal user interface: rta with no arguments. The TUI
  • UTC — Coordinated Universal Time, the zone rta time at shows an instant in beside the local one.