Observability and cost tracking for OpenWorker
See every OpenWorker session, connector action, approval and token as it happens. ClawMetry finds OpenWorker on your machine on its own and renders sessions, transcripts, cache-aware token splits and cost per session in one dashboard, next to every other agent you run.
What ClawMetry sees on OpenWorker
OpenWorker is the open source AI coworker from Andrew Ng, MIT licensed and local first. It is a desktop app rather than a coding CLI, and it is built to hand back finished work: a drafted document, a Slack reply with the numbers in it, a triaged inbox, an updated calendar. It reaches roughly 25 connected tools, Slack, GitHub, Jira, Notion, Linear, Gmail, Outlook and Google Calendar among them, alongside your files and your terminal. You bring your own model key, so one week of work can span OpenAI, Anthropic, Google and a local Ollama model. Once ClawMetry is running, that activity shows up as:
- Sessions per workspace: one row per OpenWorker session, labelled with the folder it worked in, the model behind it, the mode it ran in, when it started, when it last moved, its message count and its token totals.
- The full transcript: your request, OpenWorker's replies, its reasoning where it recorded any, and every tool call with its arguments and its result, in the order they happened. File reads, shell commands and every connector action read the same way.
- Approvals, and refusals: OpenWorker asks before it does anything consequential, sending a message, running a command, writing to a system of record. ClawMetry keeps that decision next to the action it gated: what was proposed, which connector it was aimed at, what you answered, and what never ran because you said no. Very few agents write this down at all.
- Work delegated to a team, priced: when a session staffs workers, each one is nested under the session that led it, with its own tokens and its own cost. A burst of parallel work reads as several priced children, not one unexplained jump in the parent.
- Token splits that include the cache: fresh input, cached input, cache writes and output, counted per step and added up per session, so a long autonomous run is not reported as a single call.
- Cost priced per model: because you can switch models inside a session, the dollar figure follows the model that actually produced each turn rather than whatever the session started on. A session on a local model is shown at zero rather than priced as though someone had billed you for it.
- Which connector did what: a tool call aimed at Slack, GitHub or your calendar is labelled with that connector, so you can see where a session's effort and spend actually went.
- Skills and instructions: the skills OpenWorker loads and the instructions it follows, project and global, in the same browser as every other runtime's.
Enable it in about two minutes
1. Install ClawMetry
Pick the line that matches your OS. Windows users: paste the install.cmd line, not the shell one.
# macOS or Linux
curl -fsSL https://clawmetry.com/install.sh | bash# Windows (cmd.exe or PowerShell)
curl -fsSL https://clawmetry.com/install.cmd -o install.cmd && install.cmdOr skip the terminal entirely and grab the desktop app: macOS (.dmg) · Windows (.exe) · Linux (.AppImage). It installs and updates ClawMetry for you.
2. Run the daemon
clawmetryThe daemon finds OpenWorker on the machine and starts reading immediately. No config, no restart, and nothing to add to your OpenWorker setup.
3. Open the dashboard
Open http://localhost:8900. Switch the runtime chip to OpenWorker.
Start the free trial →The one thing this catches that OpenWorker alone does not
OpenWorker is honest about the session in front of you. What no single session view can show is the shape of the week. How many sessions you left running, and in how many folders. Which ones spent their budget re-reading the same context instead of finishing the work. How much of yesterday went to delegated workers whose output you never used. Whether the model you quietly switched to last Tuesday is cheaper or just slower. Which connector is quietly the expensive one. How often you were the bottleneck, because a session sat waiting on an approval you had not seen. And how OpenWorker's total compares with the Claude Code and Codex runs happening beside it on the same laptop. ClawMetry keeps that view: cost per hour across every session and every runtime, delegated work billed separately instead of hidden inside a parent, repeated-call detection, and token-growth curves per session. Set a threshold and the approval queue pauses before the next call.
Starred by the people building AI
Engineers and founders from OpenAI, Google, PostHog and more have starred ClawMetry on GitHub. One of them helped build it.
All 27 runtimes ClawMetry observes
OpenWorker is one of 27. Free on every plan: OpenClaw, NVIDIA NemoClaw. Paid on Cloud Pro or Self-Hosted Pro: the rest.