Agent OS documentation / Reference
Ekamyth Agent OS reference
Find the command for what you want to do, then every hero command with what it does and whether it changes anything.
Find by task
Each row is a task, the command or page for it, and what you'll see.
Set up and check
| I want to | Use | What you'll see |
|---|---|---|
| Understand how Agent OS works | How Agent OS works | Your agent, Agent OS on your machine, your provider, and what you get. |
| Install Agent OS and connect my agents | Get started | One pasted line, then three checks. |
| Check that Agent OS is installed and healthy | hero status | Eight checks, each ✓ or ✗, and one next step. |
| Check that each agent is connected | hero setup --check | Each agent's connection to the model gateway. |
| Upgrade to a new version | The install line, run again | The new version checked, switched to, and a roll back: line. |
| Launch my agent so its model calls go through Agent OS | hero claude, hero codex | One launch line naming the model gateway, then your agent starts. |
| Use Agent OS from inside my agent | Commands inside your agent | The 12 /hero: commands in Claude Code and hero- skills in Codex, one line each. |
| Find the model gateway's address, or check it's up | Model gateway address | The address your agents use, and /healthz. |
| Take Agent OS out of my agents | hero setup --undo | Your agents' settings put back, and this machine's activation dropped. |
Stop and control what my agent does
| I want to | Use | What you'll see |
|---|---|---|
| Stop my agent from doing something my rules forbid | A rule in your rules file, then hero init <file> | Each rule, and whether it's checked as code. Once enforcement is on, it's refused, on Claude Code and for shell commands on Codex. |
| Stop one specific action now | Your agent's own permission settings (Claude Code permissions, Codex approval mode) | Actions on the always-refused list are refused regardless. |
| See whether each of my rules is enforced | hero enforce | Every rule Agent OS knows, where it came from, and its state on Claude Code and on Codex. |
| See what was refused, held or allowed | hero govern | Recent tool-call decisions, as ✗ refused, ⚠ needs approval or allowed. See How to read hero govern output. |
| Approve or deny an action my agent is waiting on | hero inbox | The actions held for your decision. Only you can decide, from your own terminal. |
| Know the order Agent OS checks a tool call in | The order checks run in | The order of the checks on a tool call, and routing's order. |
| Stop every model call, or switch part of Agent OS off fast | Stop or undo Agent OS | Each switch, and how to undo it. |
| See what may leave my machine | hero privacy | The declared list, which flows are on, and what was recorded as sent. |
Brain and context
| I want to | Use | What you'll see |
|---|---|---|
| Have Agent OS remember a fact | A prompt that starts "Remember that ...", or hero brain note | One line naming where the fact went and the command to undo it. |
| See what's in my Brain | hero brain status | The sections of your Brain, and when session memory was last given. |
| Check my Brain for problems | hero brain doctor | What's missing or broken, with the next step. |
| See which rules a session started with | hero memory | Which registered rules Agent OS gave each session at start. |
MCP
| I want to | Use | What you'll see |
|---|---|---|
| Understand how Agent OS works with my MCP servers | How MCP works | What Agent OS sees of your agent's MCP calls, and its limits. |
| Give my agent Agent OS's own MCP server | hero mcp | You add it to your agent's MCP configuration yourself. hero mcp never edits that configuration. |
Routing and cost
| I want to | Use | What you'll see |
|---|---|---|
| See which model answered each call | hero route | Which model served each call, and which calls moved. |
| Check which agents route through this install | hero route census | Each agent surface, and whether its model calls go through the model gateway. |
| Turn routing off | HERO_ROUTE=0 | Each call goes to the model your agent asked for. |
| See tokens by agent and session | hero usage | The token report. |
| See what my agents' work costs | hero efficiency | What the work cost, what Agent OS saved, and the levers that would lower it. |
| See every efficiency lever | hero efficiency list | Each lever, with what it found. |
| Use my own rates instead of list prices | HERO_BILLING_RATES | Cost figures at your rates. Check the file first with hero efficiency billing <path>. |
See what happened
| I want to | Use | What you'll see |
|---|---|---|
| See what my agent did in a session | hero trace | Sessions, with those that had problems first. hero trace show <id> replays one. |
| See policy violations | hero trace violations | Policy violations only, not failed steps. |
| Get one read of how my agents are doing | hero health | Findings, most severe first, each with its fix. |
| See my agents' activity, cost and checks on one screen | hero portal open | Monitor in your browser, signed in. See How Monitor works. |
| See the record of what my agents did, and check it | hero attest | The record kept on your machine, and whether it verifies. |
| See how each job ended | hero verdict | Each job's end state, judged by a watcher, not by the agent's own report. |
| Ask a question about what Agent OS recorded | hero ask | An answer, with the records it came from. |
| See or change the standing checks | hero schedule | The standing jobs, when each last ran, and its cadence. |
Fix a problem
| I want to | Use | What you'll see |
|---|---|---|
| Find out why something isn't working | Troubleshooting | Symptoms, causes and fixes, starting with hero status. |
| Get help from support | hero support | A code to email to support@ekamyth.com, with the output you saw. |
All commands
Every command with a page in these docs, in area order. "Changes" means it can change a setting, a file or a decision; see each page for exactly what.
| Command | Area | What it does | Reads or changes |
|---|---|---|---|
hero setup | Connect | Connects Claude Code and Codex to Agent OS, after showing you its plan. | --check and access read. The rest change your agents' settings. |
hero run | Connect | Launches your agent with its model calls going through the model gateway. Aliases: hero claude, hero codex. | Changes no settings file. Starts the model gateway for the run when it isn't running. |
hero proxy | Connect | Starts and manages the model gateway, which your agents' model calls pass through on your machine. | --service-status and --restart-status read. The other forms start, install, restart or remove the model gateway. |
hero status | Connect | Checks that Agent OS is installed, active and connected to your agents, and names one next step when something's wrong. | Reads only. |
hero init | Security and enforcement | Registers the rules in a rules file, and tells you which are checked as code. | Changes the rules Agent OS checks. It never edits your rules file. |
hero govern | Security and enforcement | Shows the tool-call decisions Agent OS made: refused, held or allowed. | hero govern add changes your policy file. The other subcommands are status, list, check, level, init and test; see its subcommands. |
hero enforce | Security and enforcement | Lists every rule Agent OS knows, where it came from, and whether it's enforced on Claude Code and on Codex. hero codify still works: use it to turn a rule off or on. | Reads only. |
hero inbox | Security and enforcement | Lists the actions Agent OS is holding for your decision, and approves or denies them. | approve and deny decide a hold, only from your own terminal. |
hero privacy | Security and enforcement | Shows what this install may send off your machine, and what it recorded as sent. Alias: hero audit. | Reads only. A run may send the usage heartbeat. |
hero contract | Security and enforcement | The written data contract: every outbound flow, what it carries and how to switch it off. | Reads only. |
hero codify-policy | Security and enforcement | Your switch for sending rules out for compilation: off, ask or default. | Changes that setting. |
hero brain | Brain and context | Shows, checks and adds to your Brain. | status, effective and doctor read. note, undo and init change files. |
hero brain threads | Brain and context | The commands for saving a named checkpoint of work in a workspace, and for loading a saved checkpoint again. | list, history and resume read. create and save add to your saved threads. |
hero memory | Brain and context | Shows which of your registered rules Agent OS gave your agent at session start, and simulates which would fit a budget. | Doesn't change your rules or sessions. |
hero mcp | MCP | Serves Agent OS to your agent over MCP. | See the page. |
hero route | Routing | Shows which model served each of your agents' model calls, and which calls moved. | Reads only. |
hero efficiency | Efficiency | Shows what your agent work costs, and which levers would cut it. Alias: hero cost. | Changes nothing in your work. Can send the usage heartbeat. |
hero usage | Efficiency | The token report by agent and session. | Reads only. |
hero trace | Observe | Shows what each agent session did. hero trace <id> is short for hero trace show <id>. | Reads only. |
hero attest | Observe | The record of decisions kept on your machine, and a check that it's intact. | Reads. export writes a file you name. |
hero verdict | Observe | How each job ended: done, failed or stuck, judged by a watcher, not by the agent's own report. | The bare form reads. sweep changes job states. |
hero health | Observe | Gives one read of how your agents are doing, with a fix for each finding. | Reads your records. --html writes a report file and opens it. |
hero ask | Observe | Answers a question from what Agent OS recorded, with the records it came from. | Reads your records. |
hero schedule | Observe | Lists the standing jobs Agent OS runs for you on a cadence, and lets you change a job's cadence or switch it on or off. | list and --service-status read. The other forms change something. |
Interface: hero portal opens the Portal, whose main screen is Monitor.
Other surfaces
The customer surfaces are the hero commands, the model gateway's local HTTP endpoints, the hooks hero setup installs, MCP, configuration and data files, skills and customer settings. Ekamyth's own server endpoints and the Portal's HTTP routes are not a supported API.
- Model gateway address: the address and port your agents use, the model gateway's HTTP endpoints, the health check, and its environment variables.
- Settings and environment variables: the files you edit and the variables you set.
- Data files: what Agent OS keeps on your machine, and why.
- Commands inside your agent: the slash commands and skills setup installs.
Checked against Agent OS 0.1.36rc14 (build 599c5e120c18) on 17 Nov 2026.