Skip to main content
Docs / CLI Reference

The agentomy-agent CLI.

agentomy-agent is a command-line governance peer: drop it into any agent session to get a governance system prompt, a model-capability check, and a live governance-state readout. It has four subcommands: init, session, status, and serve. Every flag and default below is taken directly from the source.

Install and invoke

agentomy-agent ships as a Python package (Python 3.9 or newer). Install it with pip, then invoke it as agentomy-agent. Running it with no subcommand prints the built-in help.

# Install the CLI (Python 3.9 or newer)
$ pip install agentomy-agent
# Confirm the install
$ agentomy-agent --version
# Print built-in help
$ agentomy-agent

Two acronyms show up below. LLM means large language model (the model running your agent, such as a local Ollama model or a cloud model like Claude). MCP means Model Context Protocol, an open standard for exposing tools to an LLM runtime. The serve command speaks it so local runtimes like Ollama can pull the governance prompt as a tool.

agentomy-agent initSubcommand

Detect local LLMs and install the governance prompt. Scans for local model endpoints (Ollama on localhost:11434 and LM Studio on localhost:1234), reports the models it finds with size tags and a recommendation (32B parameters or larger is recommended), then prints the governance system prompt for you to paste into your LLM session.

Syntax

$ agentomy-agent init [--full-prompt]

Flags

FlagMeaningDefault
--full-promptPrint the complete governance system prompt after the summary.off (first 2000 characters shown)

Example

# Detect local models and print the governance prompt
$ agentomy-agent init
# Print the full, untruncated system prompt
$ agentomy-agent init --full-prompt

agentomy-agent sessionSubcommand

Start a governed session. Prints a model-capability assessment for the model you name, the governance context for the session, and the system prompt to paste into your LLM. If you pass an infrastructure URL and it is reachable, the session runs in connected mode (6/6); otherwise it runs in standalone embedded governance mode (3/6).

Syntax

$ agentomy-agent session [--model MODEL] [--scope SCOPE] [--infra URL] [--full-prompt]

Flags

FlagMeaningDefault
--model MODELModel name or identifier used for the capability assessment (e.g. qwen2.5:32b, claude).unknown
--scope SCOPEFree-text description of what the session is authorized to do (e.g. security-research).general
--infra URLAgentomy infrastructure base URL. If reachable, the session upgrades to connected 6/6 mode; if not, it stays standalone 3/6.none (standalone)
--full-promptPrint the complete system prompt instead of a truncated preview.off (first 1500 characters shown)

Example

# Start a governed session on a local 32B model
$ agentomy-agent session --model qwen2.5:32b --scope security-research
# Check for connected 6/6 mode against a local Agentomy instance
$ agentomy-agent session --model claude --infra http://localhost:3000

agentomy-agent statusSubcommand

Show the current governance state. With no infrastructure URL it reports Standalone mode: 3 of the 6 GovernanceBench dimensions are active (Authorization and Behavioral Integrity). Pass an infrastructure URL and, if it is reachable, status reports Connected mode: all 6 dimensions, plus fleet health, quarantined-agent count, audit-trail integrity, and anomaly-detector state.

Syntax

$ agentomy-agent status [--infra URL]

Flags

FlagMeaningDefault
--infra URLAgentomy infrastructure base URL to check for connected mode (e.g. http://localhost:3000).none (standalone)

Example

# Show standalone 3/6 status
$ agentomy-agent status
# Check connected 6/6 status against local infrastructure
$ agentomy-agent status --infra http://localhost:3000

agentomy-agent serveSubcommand

Start MCP server mode for Ollama integration. Runs a small HTTP server that exposes the governance prompt and status as MCP (Model Context Protocol) tools, so a local runtime like Ollama can pull them in as tools. It serves GET /health, GET /prompt, GET /mcp/tools, and POST /mcp/tools, and prints a ready-to-paste Ollama MCP config. Two tools are exposed: agentomy_prompt (returns the governance system prompt) and agentomy_status (returns local standalone governance status). Press Ctrl+C to stop.

Syntax

$ agentomy-agent serve [--host HOST] [--port PORT]

Flags

FlagMeaningDefault
--host HOSTNetwork interface to bind the server to.127.0.0.1
--port PORTTCP port to listen on.8765

Example

# Serve on the default 127.0.0.1:8765
$ agentomy-agent serve
# Bind all interfaces on a custom port
$ agentomy-agent serve --host 0.0.0.0 --port 9000

Want the meaning behind 3/6 and 6/6, standalone vs. connected mode, and the operator tiers? Read the concepts guide.

Core Concepts