systemprompt-cli

One binary to run the infrastructure you own: agent orchestration, MCP governance, analytics, profiles, cloud deployment, and self-hosted operations. Every command runs two ways, an interactive mode for humans and a non-interactive mode for agents, over the same code path.

Layer: Entry — application boundary. Binary: systemprompt. Library: systemprompt_cli. Part of the systemprompt-core workspace.

Overview

The CLI exposes systemprompt.io through eight top-level domains:

Domain	Purpose
`core`	Skills, content, files, contexts, plugins, hooks, artefacts
`infra`	Service lifecycle, database, jobs, log streaming
`admin`	Users, agents, configuration, session, setup wizard, bridge enrolment, access-control baseline
`cloud`	Authentication, tenants, profiles, deploy, sync, secrets, custom domains, Dockerfile, database
`analytics`	Overview, conversations, agents, tools, requests, sessions, content, traffic, costs
`web`	Content types, templates, assets, sitemap, validation
`plugins`	Extension discovery, configuration, execution, capability inspection, MCP server management
`build`	Build core workspace, build MCP extensions

Architecture

Top-level modules. docs.rs carries the file-level detail.

src/
├── lib.rs          # Entry point: pub async fn run(), command routing, output finalisation
├── descriptor.rs   # CommandDescriptor: declares initialisation needs per command
├── cli_settings.rs # CliConfig, OutputFormat, VerbosityLevel
├── interactive.rs  # Interactive menu mode
├── runner/         # args (clap Commands enum), bootstrap, profile routing, structured output
├── session/        # Session lifecycle (JWT, context, persistence)
├── presentation/   # Output rendering: tables, JSON, YAML, widgets
├── shared/         # Cross-cutting utilities (parsers, paths, docker, profile)
└── commands/       # One module per domain (see the domain table above)

The top-level Commands enum in src/runner/args.rs dispatches to per-domain *Commands subcommand enums declared in each commands/<domain>/mod.rs.

Core Modules

Module	Purpose
`lib.rs`	`pub async fn run()` entry point, top-level routing, output finalisation
`runner/args.rs`	clap-derived `Cli` and `Commands` definitions
`runner/bootstrap.rs`	Initialisation sequence: profile → credentials → secrets → paths → validation
`cli_settings.rs`	`CliConfig`, `OutputFormat`, `VerbosityLevel`
`descriptor.rs`	`CommandDescriptor` constants (`NONE`, `PROFILE_ONLY`, `PROFILE_AND_SECRETS`, `PROFILE_SECRETS_AND_PATHS`, `FULL`) declared per command
`runner/routing/`	`ExecutionTarget` (local vs remote SSE streaming)
`session/`	JWT session tokens, active context, on-disk persistence
`presentation/`	Format-aware renderer, render state, terminal widgets
`shared/`	`CommandResult<T>`, docker utilities, value parsers, process helpers, profile/project detection

Command Requirements

Each command variant declares a CommandDescriptor indicating what bootstrap state it needs:

NONE — standalone (no profile, no secrets, no database)
PROFILE_ONLY — profile loaded
PROFILE_AND_SECRETS — profile and secrets loaded
PROFILE_SECRETS_AND_PATHS — profile, secrets, and resolved paths
FULL — profile, secrets, paths, validation, database pool

runner/bootstrap.rs walks the descriptor and short-circuits unneeded steps.

Output System

All commands return CommandResult<T>:

CommandResult::table(data)
    .with_title("Title")
    .with_hints(json!({ "columns": [...] }))

Artefact variants: Table, List, Card, Text, CopyPasteText, Chart, Form, Dashboard. The renderer picks a representation based on --json, --yaml, or interactive TTY.

Usage

[dependencies]
systemprompt-cli = "0.21"

cargo install systemprompt-cli

Dual-Mode Operation

Every command supports two modes:

Mode	Audience	Behaviour
Interactive	Humans	Prompts, confirmations, coloured output
Non-interactive	Agents	All inputs via flags, structured output, no prompts

# Interactive
systemprompt admin agents create

# Non-interactive
systemprompt --non-interactive --json admin agents create --name myagent

Standard Flags

Flag	Short	Purpose
`--yes`	`-y`	Skip confirmation
`--dry-run`		Preview without executing
`--force`		Override safety checks
`--json`		JSON output
`--yaml`		YAML output
`--non-interactive`		Disable prompts
`--quiet`		Minimal output
`--verbose`		Detailed output

License

BSL-1.1 (Business Source License). Source-available for evaluation, testing, and non-production use. Production use requires a commercial licence. Each version converts to Apache 2.0 four years after publication. See LICENSE.

systemprompt.io · Documentation · Guides · Live Demo · Template · crates.io · docs.rs · Discord

Entry layer · Own how your organisation uses AI.

systemprompt-cli 0.21.1

Production infrastructure for AI agents