Ekamyth Agent OS docs
Get started / Things to try

Things to try with Ekamyth Agent OS

Eight short tries that show Agent OS at work on your own machine. Each one works on a fresh install, once you've finished Get started.

Before you start

Each try shows a short sample of what to expect. The samples come from a test machine with simulated data, so your counts, models and times differ. Lines are left out where you see …

1. Check that everything is running

What it shows: whether Agent OS is installed, active and connected: the model gateway, your agents, the hooks and enforcement, each with a ✓ or ✗ and one next step.

Run:

hero status

Sample output (simulated data). On this machine the agents weren't yet pointed at the model gateway, so the next: line names hero setup.

status — Ekamyth on this machine
  8 checks · 7 ok · 1 needs attention
  ┌───┬───────────────┬────────────────────────────────────────────────────┐
  │   │ check         │ state                                              │
  ├───┼───────────────┼────────────────────────────────────────────────────┤
  │ ✓ │ install       │ ekamyth 0.1.36rc13 (build cd08bd39c026)            │
  │ ✓ │ access        │ active                                             │
  │ ✓ │ proxy         │ running                                            │
  │ ✗ │ agents        │ Claude Code, Codex not routed through the proxy    │
  │ ✓ │ hooks         │ installed and approved                             │
  │ ✓ │ enforcement   │ armed                                              │
  │ ✓ │ stores        │ readable                                           │
  │ ✓ │ disk          │ enough free space · stores within limits           │
  └───┴───────────────┴────────────────────────────────────────────────────┘
  rules  8 registered · 2 blocked
  next: hero setup
  paths, ports and sizes: hero status --details

The proxy row is the model gateway. Run the command after next:, then hero status again. See How to read hero status output.

2. See Agent OS stop a risky action

What it shows: a real decision at the hook. A recursive forced delete is on the list Agent OS always refuses, so the hook refuses it before it runs, on Claude Code and Codex alike.

Do this: in a project folder, make an empty folder to try it on:

mkdir ekamyth-try

Then ask your agent, in a new Claude Code or Codex session:

Run rm -rf ekamyth-try

Your agent reports that the command was refused, and the folder is still there. Then see the refusal in the record:

hero trace list

Sample output (simulated data). A session with one refused step.

  ┌───────────────────┬──────────────┬────────┬──────────────────────────┬────────────┬────────────┐
  │ when (UTC)        │ session      │ agent  │ steps                    │ outcome    │ tokens ·   │
  │                   │              │        │                          │            │ cost       │
  ├───────────────────┼──────────────┼────────┼──────────────────────────┼────────────┼────────────┤
  │ 11-07 06:11–06:53 │ claude-3cd53 │ claude │ 11 tools (Edit, Grep,    │ ✗ 1        │ 17,197 ·   │
  …

To remove the empty folder, run rmdir ekamyth-try in your own terminal. For the whole list, see Always refused. To add a rule of your own, see hero init.

3. Read what your agents did

What it shows: one row per session, with the session id and the agent that ran it, its tool calls and model calls, and how it ended.

Run:

hero trace list

Sample output (simulated data).

Trace · newest 25 sessions · rolling 30d (Oct 09–Nov 08 UTC)
  …
  ┌───────────────────┬──────────────┬────────┬──────────────────────────┬────────────┬────────────┐
  │ when (UTC)        │ session      │ agent  │ steps                    │ outcome    │ tokens ·   │
  │                   │              │        │                          │            │ cost       │
  ├───────────────────┼──────────────┼────────┼──────────────────────────┼────────────┼────────────┤
  │ 11-08 14:23       │ claude-3cd63 │ claude │ 1 tool (Bash) · 1 model  │ ✓ ok       │ 20,200 ·   │
  …
  ├───────────────────┼──────────────┼────────┼──────────────────────────┼────────────┼────────────┤
  │ 11-08 10:38–11:26 │ codex-3cd61  │ codex  │ 5 tools (Read, Grep,     │ ✓ ok       │ 9,335 ·    │
  …

See hero trace list.

4. Replay one session step by step

What it shows: what happened inside one session, in order: the model request, the route it took, the governance decision, each tool call and the outcome.

Run, with a session id from the session column of try 3:

hero trace show <session>

Sample output (simulated data).

  TRACE  session=claude-3cd63  ·  2 step(s)  ·  ✓ OK  ·  2026-11-08T14:23:06Z
  ───────────────────────────────────────────────────────────────────────────
  actor proxy  ·  model claude-sonnet-5  ·  elapsed 25.55s

  step  t+            dur  stage    recorded
     1  +0.000s    25.55s  request  provider anthropic
     1  +0.000s            route    model claude-sonnet-5 · class code
     1  +0.000s            govern   ✓ allow · rule not attributed
     2  +0.000s    40.0ms  tool     agent:Bash ran
  …
     1  +0.000s    25.55s  outcome  ✓ ok · surface claude
     2  +0.000s    40.0ms  outcome  ✓ ok · surface claude · agent:Bash
  …

See Example 1 of hero trace.

5. See how your model calls were routed

What it shows: each recent model call, the kind of work it was, and the model that served it. A call may move to another model; (kept) means it stayed on the model your agent asked for.

Run:

hero route

Sample output (simulated data).

Route · local Runtime scope
  last 24h · partial
RECENT · each row is one metered record, not a distinct call
  06:23 · Model call · code · routed to claude-sonnet-5 (kept) · —
  05:46 · Model call · review · routed to claude-haiku-4-5 (kept) · —
  …

IS IT WORKING
  routing ON
  Codex agentic floor: off — native Codex keeps the requested model (HERO_ROUTE_CODEX_AGENTIC_FLOOR_TIER is not set on the Runtime)
  …
  0 of 10 metered records changed model (record count, not distinct calls)
  …

MODELS BY TASK CLASS
  classify · claude-sonnet-5: 1 records
  classify · gpt-5.3-codex: 1 records
  code · claude-sonnet-5: 2 records
  …

To check which of your agents send their calls through Agent OS at all, run hero route census. See How to read hero route output.

6. See what the work cost, and where the waste is

What it shows: the calls in the last 30 days, the levers Efficiency checks, which moves routing made, and the waste your agent could change.

Run:

hero efficiency

Sample output (simulated data). Lines with dollar amounts are left out where you see …

Efficiency · rolling 30d (Oct 09–Nov 08 UTC) · 311 calls
  …
  311 metered records (incl. 0 evaluation replays)
  Ekamyth Agent OS saved       unknown not measured — the price of the model you
                                       asked for wasn't recorded (7 buckets not
                                       measured yet)
  …

TOP MOVES
  newest 50 of 311 records read · counts cover only these
  gpt-5.4 → gpt-5.4-mini · 3 moved records · cost-min
  claude-opus-4-8 → claude-sonnet-5 · 1 moved record · cost-min

NOT SAVED YET  6 levers with an opportunity · 3 not priced — no price for this
               lever's evidence · your agent could change: blind retries, long
               outputs, repeated tool calls, calls still on the expensive model
               · --details
  …

See Example output of hero efficiency.

7. Get one read of how your agents are doing

What it shows: findings across Govern, Optimize, Assure and Ledger, most severe first, each with a fix. To check whether Agent OS's own parts are running, use try 1.

Run:

hero health

Sample output (simulated data).

Agent health · rolling 30d (Oct 09–Nov 08 UTC)
  …
  0 of 4 areas failing · 9 findings · govern builder — records all calls; stops a short list of
    catastrophic actions
  ┌──────────┬─────┬──────────────────────────────────────────────┬────────────────────────────────┐
  │ area     │ sev │ finding                                      │ fix                            │
  ├──────────┼─────┼──────────────────────────────────────────────┼────────────────────────────────┤
  …
  │ Assure   │ MED │ 21 of 311 proxied calls failed in this       │ hero help trace                │
  │          │     │ window (error answer or refused before       │                                │
  │          │     │ dispatch)                                    │                                │
  …
  └──────────┴─────┴──────────────────────────────────────────────┴────────────────────────────────┘
  Run: hero help route
  Govern ✓ · Optimize ! · Assure ! · Ledger ✓ · evidence and read provenance: hero health --details

See How to read hero health output.

8. See what can leave your machine

What it shows: every flow Agent OS declares it may send to Ekamyth, what each carries and whether it's on, then the sends its egress guard recorded.

Run:

hero privacy

Sample output (simulated data). In the run that printed this, the usage beacon was turned off.

Privacy · what this install may send out, and what its egress guard recorded
WHAT MAY LEAVE — the declared contract
  …
  ┌─────────────────────┬──────────────────┬─────────────────┬─────────────────┐
  │ flow                │ to               │ carries         │ when            │
  ├─────────────────────┼──────────────────┼─────────────────┼─────────────────┤
  …
  │ usage-beacon        │ core.ekamyth.com │ anonymous       │ off · turn off: │
  │                     │                  │ install         │ HERO_BEACON=off │
  │                     │                  │ heartbeat       │                 │
  │                     │                  │ (random install │                 │
  │                     │                  │ id + version +  │                 │
  │                     │                  │ platform +      │                 │
  │                     │                  │ event)          │                 │
  …
  └─────────────────────┴──────────────────┴─────────────────┴─────────────────┘
  NOT RECORDED = this flow sends without the egress guard today: its destination
    is not checked and its sends never appear below (4 of 11 flows)

WHAT ACTUALLY LEFT — as recorded by the egress guard
  the guard has recorded no sends — every send through it is logged here
  not a full record: rule-compile, decision-metadata, usage-metrics, policy-pull
    send without the guard and are never listed here

See How Privacy works.

Next steps

Checked against Agent OS 0.1.36rc14 (build 599c5e120c18) on 17 Nov 2026.