How Do I Track AI Agents Across Multiple Machines?
Track AI agents on a laptop, a server and a VPS in one timeline: one API key per machine, and consistent agent and project names.
Reviewed by WaveBrain Operations TeamReviewed 5 min read
The short answer
To track AI agents across multiple machines, have every agent on every machine report its actions to one shared log instead of a local file. WaveBrain gives you one timeline for all of them: create one API key per machine, use the same agent and project names everywhere, and filter by agent or project to see what ran where.
You run Claude Code on your laptop during the day, a Hermes bot on a home server that answers support email, and a cron job on a small VPS that posts to social accounts. On Monday morning a customer says they got two invoices. To find out which agent sent them you SSH into the server, scroll through a Hermes log, check the VPS cron output, then open your laptop's Claude Code history. Twenty minutes later you know, and you still aren't sure nothing else happened.
Every agent keeps its own record on the machine it runs on. Tracking them across machines comes down to sending all of those records to one place, with names that let you tell them apart.
How do I track AI agents running on different machines?#
Point every agent at the same activity log and give each machine its own credentials. With WaveBrain, agents report what they did through the MCP server at https://getwavebrain.com/api/mcp or the REST endpoint at https://getwavebrain.com/api/v1/actions. Every entry lands in one searchable timeline, which you filter by agent, project and action type, or search by title and details.
The setup per machine is short:
- Create an API key for that machine in the WaveBrain dashboard.
- Connect the agents on it (MCP config for Claude Code, Codex, Cursor or Hermes, a REST call for scripts).
- Add the logging rules to each agent's instructions file so it reports actions without being asked.
- Use one naming scheme for agents and projects on every machine.
Why is it hard to see what agents did on a laptop, server and VPS?#
Each agent writes its history to the machine it runs on, in its own format. Claude Code keeps transcripts on your laptop, Hermes keeps sessions and state inside its own home directory (the Hermes docs list config.yaml, .env, memories, sessions and a state database per profile), and a cron script writes wherever you pointed its output. None of these talk to each other, and a VPS you rarely log into is the one most likely to hold the surprise.
Machines also sleep and restart. Your laptop is closed at night, which is exactly when the VPS cron job runs, so whatever you use to watch them should live somewhere that stays up.
Should each machine get its own API key?#
Yes. One key per machine lets you revoke a single machine's access without touching the others, which matters when a VPS gets rebuilt or a laptop is lost. The key form in the dashboard suggests names in the style of "Laptop Claude Code", so name keys by machine and agent, for example vps-social-cron or homeserver-hermes.
The Free plan includes 2 API keys, enough for a laptop and one server. If you run more machines than that, agents can share a key, and you tell them apart with the agent name instead. Pro is coming with 50 keys. API keys are stored as hashes, so copy a new key when it's shown.
How should I name agents and projects?#
Name agents after the job and the machine, and name projects after the thing the work affects. Agent names are stored lowercase, so pick lowercase names from the start to avoid surprises. Keep the same project name on every machine so a filter shows all work on that project regardless of where it ran.
| Machine | Agent | Agent name | Project | Connects through |
|---|---|---|---|---|
| Laptop | Claude Code | claude-code-laptop | northwind/web | MCP plus plugin hooks |
| Home server | Hermes support bot | hermes-support | northwind/support | MCP in config.yaml |
| VPS | Social posting cron job | social-cron-vps | northwind/marketing | REST call |
| GitHub Actions | Deploy workflow | ci-deploy | northwind/web | REST call |
With this scheme, filtering by northwind/web shows the laptop's commits next to the CI deploys, and filtering by hermes-support shows only what the support bot did. A project in the owner/area form, like northwind/growth on the WaveBrain homepage, stays readable as the list grows.
How do I connect the agents on each machine?#
Use the config for the client on that machine; the API key is the only thing that changes. On a laptop with Claude Code, run the command from the Setup page:
claude mcp add --transport http --scope user wavebrain https://getwavebrain.com/api/mcp \
--header "Authorization: Bearer <YOUR_API_KEY>"
On a server running Hermes, add WaveBrain under mcp_servers in ~/.hermes/config.yaml and put the key in ~/.hermes/.env (the Hermes setup guide has the exact block). On a VPS that only runs scripts, call the REST endpoint at the end of each job:
curl -X POST https://getwavebrain.com/api/v1/actions \
-H "Authorization: Bearer $WAVEBRAIN_API_KEY" \
-H "Content-Type: application/json" \
-d '{"agent": "social-cron-vps", "type": "publish", "title": "Posted the weekly product update to LinkedIn", "project": "northwind/marketing", "status": "success"}'
For Claude Code and Codex there are also open source plugins with hooks that catch high-stakes actions (deploys, pushes, publishes, payments, emails, database writes) even when the agent forgets to log them.
What happens when a machine goes offline or WaveBrain is unreachable?#
The agents keep working. If WaveBrain can't be reached, log_action returns an error and the task continues. With the Claude Code or Codex plugin, hook entries wait in a local queue at ~/.wavebrain/queue and are retried on the next turn.
WaveBrain is not an uptime monitor, though. It records actions that agents report, so a machine that is switched off produces no entries rather than an alert. If you need to know when a server goes down, keep a separate health check for that. To get a message when an agent does report a failure, connect Slack (see Slack alerts for agent failures).
If you're juggling agents on more than one machine, start free. The free plan covers 1,000 actions a month, 7 days of history and 2 API keys, with no credit card.
Frequently asked questions#
Can I use one API key on all my machines?#
Yes, it works, but separate keys let you revoke one machine without breaking the others. If you share a key, give each agent a distinct name so the timeline still shows where each action came from.
How do I tell which machine an action came from?#
Put the machine in the agent name, such as hermes-homeserver, or add it to the metadata field. You can then filter the timeline by agent.
Do I need to install anything on the VPS?#
Not for scripts. A cron job only needs to send an HTTP POST to the REST endpoint with your API key, which curl already does.
Does WaveBrain read logs from my servers?#
No. It never reads your repo or files and stores only what agents and hooks send, such as the title, details, project, link and status.
How long is history kept across machines?#
The Free plan keeps 7 days of history and Pro, which is coming, keeps 90 days. Older entries are purged daily for every machine alike.