cli.md

CLI guide

The activlayer command is the Community Edition control plane. It discovers .activlayer in the current directory or a parent directory. Override discovery with ACTIVLAYER_HOME or --home.

Command map

activlayer
├── init                    create a one-organization environment
├── status                  environment dashboard
├── doctor                  configuration and connectivity checks
├── guide                   guided operational workflow
├── serve                   authenticated HTTP API
├── config show|get|set     environment properties
├── user list|add|remove|token
├── llm list|select|add|use|test|remove
├── connector list|add|test
├── extension list|add|remove
├── chat                    plan and review natural-language design changes
├── knowledge collection-create|collection-list|add|search
├── memory list|search      inspect one worker's isolated memory
├── agent new|list|show|set|workers|validate|import|export|provision|publish|graph
│   └── node list|types|explain|show|add|set|remove|connect|disconnect
└── run start|list|show|resume|approve

Every group and command supports --help.

activlayer init automatically probes the standard loopback endpoints for Ollama, vLLM, and llama.cpp, plus non-default ports declared by running vLLM and llama.cpp processes. Discovered servers with at least one model are saved and the first becomes active. Pass --no-detect-llm for isolated or manually configured environments.

activlayer llm list repeats live discovery and merges it with configured local or remote providers. Use activlayer llm select <displayed-name> to configure a discovered provider and make it active. llm use <name> remains available for switching between already configured providers.

Agent design workflow

Create a definition, browse available node types, add nodes, edit nested configuration, connect the graph, validate, and publish:

activlayer agent new "Content Review" --id content-review
activlayer agent node types --category ai
activlayer agent node explain control.approval
activlayer agent node add content-review analyze --type ai.prompt
activlayer agent node set content-review analyze data.config.prompt \
  "Review the submitted content: {content}"
activlayer agent node connect content-review trigger analyze
activlayer agent graph content-review
activlayer agent validate content-review
activlayer agent publish content-review

agent set changes any non-graph agent property. agent node set changes any node property. Dotted paths create missing intermediate objects. Values are decoded as JSON when valid; use a JSON-quoted string when a value might otherwise be interpreted as a number, boolean, array, or object.

Portable definitions

agent import installs a JSON file as a draft. agent provision combines validation and install, and --publish creates an execution snapshot. agent export creates a portable file that can be opened by another Community environment or Studio.

Shell completion

activlayer --install-completion

Typer installs completion for the active shell.

Multi-agent design

Create workers before their orchestrator, publish the workers first, and then publish the orchestrator:

activlayer agent new "Product Worker" --id product-worker --type worker
activlayer agent publish product-worker
activlayer agent new "Service Orchestrator" --id service-orchestrator --type orchestrator
activlayer agent workers service-orchestrator product-worker
activlayer agent publish service-orchestrator

agent validate checks role relationships as well as graph structure. Publication fails when an orchestrator references a missing, unpublished, or non-worker agent.

Use activlayer chat "<request>" to produce a validated design plan. The command displays every typed operation and asks before changing drafts. --apply is intended for automation where the request and generated plan are already subject to review.