Observability and cost tracking for Devin
See every Devin CLI session, tool call, subagent and token as it happens. ClawMetry detects the Devin CLI on your machine automatically and renders sessions, transcripts, per-turn token and cache usage, response timing and subagent fan-out in one dashboard, next to every other agent you run.
What ClawMetry sees on Devin
Devin is Cognition's autonomous software engineer. This page covers Devin CLI, the local
one: you install it from cli.devin.ai,
type devin inside a project, and it works on your files with your shell on the frontier model
you picked, handing the job to a Devin cloud agent when the work outgrows your laptop. Once ClawMetry is
running, that activity shows up as:
- Sessions per conversation: one row per Devin CLI session, with the directory it ran in, the model, the mode it ran under, started at, last activity, message count, and per-session token totals.
- Live activity: every tool call and its result as it lands, with arguments, plus the model's thinking where Devin recorded it. A single call is shown once, not once per place Devin mentions it.
- Real token and cache usage: the input, output, cache-read and cache-write tokens Devin records for each turn, with time-to-first-token and response duration, totalled per session. No dollar figure and no ACU figure: Devin bills in agent compute units and does not keep them on your machine, so both would be invented.
- Subagent fan-out: the turn that delegated, the brief it handed over, and the answer that came back, so background work is a step you can open instead of a sentence of summary.
- Reverted work, counted once: rewinding or forking leaves the abandoned turns behind in the session. ClawMetry follows the live chain and reports the abandoned turns separately, so a session you rewound twice does not read as triple the spend.
- Skills and rules: the skills and always-on rules Devin loads, global and per project, in the same browser as every other runtime's.
- Next to everything else: Devin sits in the same dashboard as Claude Code, Codex, Cursor and the rest, so fleet cost per hour is one number instead of six tabs.
Subagents stop being a black box
Devin CLI delegates focused work to subagents with its run_subagent tool, either because you
asked for it or because it decided a task was worth doing independently. Cognition's own documentation is
blunt about what you get back: you do not see the subagent's raw output, the parent reads the result and
summarises the key findings for you. In background mode the parent keeps working while the child runs, so
the summary is often the only trace that a second agent existed at all.
ClawMetry keeps the fan-out visible. The delegating turn, the brief the child was handed, and the reply that came back are all on the session timeline, in the order they happened, alongside the tool calls the parent ran itself. When a session took three times as long as you expected, you can see which branch of the work it went into rather than reading a summary of a summary.
- Devin Cloud (app.devin.ai): hosted sessions exist only behind Devin's API. Reading them would mean giving ClawMetry an enterprise API key and letting the daemon call a vendor on your behalf. That belongs in a connector you switch on deliberately, not in a background read path. Hand a session off to a cloud agent and ClawMetry stops where your machine stops.
- Devin Desktop (the IDE, formerly Windsurf): its Cascade sessions are not written to disk at all, so there is nothing on your machine to read. We checked, with real turns.
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.cmd2. Run the daemon
clawmetryThe daemon auto-detects the Devin CLI on the machine and starts reading immediately. No config, no runtime restart, nothing to add to your Devin setup.
3. Open the dashboard
Open http://localhost:8900. Switch the runtime chip to Devin.
Start the free trial →The one thing this catches that Devin alone does not
Inside a Devin session you can see that session. /session-stats tells you what this
conversation cost. What neither shows you is the shape of the week: how many sessions you left running in
how many repos, which ones burned tokens re-reading the same files, how many of yesterday's turns went into
branches you rewound and threw away, how much of the work happened inside subagents you only ever saw
summarised, and how Devin's total compares with the Claude Code and Codex runs happening beside it.
ClawMetry keeps that view: spend per hour across every session and every runtime, abandoned branches counted
separately from the work you kept, repeated-tool-call detection, and per-session token-growth curves. Set a
threshold and the approval queue pauses before the next call. Kill switch (Pro) stops it.
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 26 runtimes ClawMetry observes
Devin is one of 26. Free on every plan: OpenClaw, NVIDIA NemoClaw, Goose. Paid on Cloud Pro or Self-Hosted Pro: the rest.