Managed agents are available on Enterprise. Signal is available on all Arize AX plans. See Pricing and usage for plan limits.
ag CLI is the terminal and TUI client for managed agents and Signal. It talks to the Arize GraphQL API with your API key and a space, and lets you:
- Run and follow managed agent sessions (one sandboxed agent run each)
- Manage presets (reusable agent configs) and skills (credentials injected into the sandbox)
- Enable Signal on a project, browse its issues, and trigger a Signal run now
- Open an interactive TUI by running
agwith no arguments
ag is not the ax CLI. Use ax for platform resources (spaces, projects, traces, datasets, evaluators, prompts). Use ag for the agent runtime: sessions, presets, skills, automations, and Signal. The two tools accept the same Arize API key.Getting Started
1. Install the CLI
Requires Go on your
PATH. The binary installs to ~/go/bin; keep that directory on your PATH.Prebuilt binaries and a Homebrew formula are not published yet. Contact your Arize account team if you need access to the source repository.
2. Save a context
Every space-scoped command needs a space ID (the/spaces/<id> segment of any Arize URL) and an API key. Save both once so you do not have to pass them on every command:
--api-key-env stores the name of the environment variable, never the key itself. Run ag context show to confirm what was saved.
3. Verify your setup
Configuration
Context file
ag context set writes ~/.arize/ag_context.toml with three values: the default space, the GraphQL endpoint, and the name of the environment variable that holds your API key. Flags passed on the command line override the saved values for that invocation.
Authentication
ag looks for an API key in this order and uses the first one it finds:
--api-key <key>on the command--api-key-env <VAR>on the command or in the saved context- The
ARIZE_API_KEYenvironment variable - The
axprofile named by--profile, or the activeaxprofile
Endpoint
The GraphQL endpoint is resolved in this order:--endpointon the command, or the saved context- The
axprofile’s GraphQL routing, when--profileis used - The
ARIZE_GRAPHQL_ENDPOINTenvironment variable https://app.arize.com/graphql
--endpoint to their own host. The path must be empty or /graphql.
Options & Output
Common Options
Every command exceptag context accepts these options:
--space is accepted by the commands that are scoped to a space: session list, session create, signal project list, automation list, and the TUI. It falls back to the saved context.
Output Formats
With--output text, list commands print a table and accept --columns to choose and order the columns. With --output json, they print the full objects, including fields that have no column.
IDs
Every ID thatag prints or accepts is a relay ID (a base64 string), not a numeric database ID. Copy IDs from ag ... list output or from the Arize URL.
Aliases
Resource names have aliases, so these are the same command:ag session=ag sessions=ag jobsag preset=ag presets=ag agentsag skill=ag skills=ag integration=ag integrationsag automation=ag automationsag signal=ag signals
session, preset, and skill.
Interactive TUI
Runag with no subcommand to open a home menu with two areas: Sessions (create and inspect sessions) and Signal (browse Signal issues for a project). Press esc to return to the home menu and q to quit.
Command cheat sheet
Resources
Sessions
Create, message, follow, and inspect managed agent runs
Presets
Reusable agent configs: repo, provider, and skills
Skills
Credentials injected into the sandbox as environment variables
Automations
List scheduled runs and trigger one now
Signal
Enable Signal on a project and browse its issues
GraphQL guide
The same operations from your own code
Related docs
- Managed agents — The gallery, agent configs, sessions, and automations in the UI
- Connect your harness — Credentials, skills, permissions, and sandbox boundaries
- Signal — Ranked issues on a tracing project
- AX CLI — The
axCLI for platform resources