Which repository is actually expensive?
Spend rolls up per repository, so you know which project is driving the invoice before it lands.
- hashtrack-cloud
- api-gateway
- sync-worker
- docs-site
The ledger for AI-assisted engineering.
Hashtrack captures every agent session across Claude Code, OpenCode, and pi — as a plugin or extension inside the tools you already use — then rolls the tokens up by repository, model, and developer, so engineering leads can price AI work instead of guessing at the invoice.
See every token, by user
Hashtrack breaks token usage down per person, so you know who is using their budget.
AL
Alice BO
Bob CA
Carol DA
DaveWhich models are doing the work
Every token is attributed to the model that spent it, so you can see which models perform best.
What your team leans on
Skills ranked by how often your team runs them, across every harness.
Any agent, one source of truth
Hashtrack pulls token and cost data out of every harness your team runs.
What each repository costs
Every dollar is attributed to the repo that spent it, so each project carries its own budget line.
Every harness reports into the same ledger
Hashtrack reads token and cost data from the tools your team already runs. Nothing new to adopt, nothing to migrate.
Claude Code 
GitHub Copilot soon
Codex soon
Cursor soon
%20{%20:root%20{%20filter:%20none;%20}%20}%20@media%20(prefers-color-scheme:%20dark)%20{%20:root%20{%20filter:%20none;%20}%20}%20%3c/style%3e%3c/svg%3e)
OpenCode 
pi - Claude Code
- GitHub Copilot soon
- Codex soon
- Cursor soon
- OpenCode
- pi
- Claude Code
- GitHub Copilot soon
- Codex soon
- Cursor soon
- OpenCode
- pi
- Claude Code
- GitHub Copilot soon
- Codex soon
- Cursor soon
- OpenCode
- pi
- Claude Code
- GitHub Copilot soon
- Codex soon
- Cursor soon
- OpenCode
- pi
- Claude Code
- GitHub Copilot soon
- Codex soon
- Cursor soon
- OpenCode
- pi
- Claude Code
- GitHub Copilot soon
- Codex soon
- Cursor soon
- OpenCode
- pi
- Claude Code
- GitHub Copilot soon
- Codex soon
- Cursor soon
- OpenCode
- pi
- Claude Code
- GitHub Copilot soon
- Codex soon
- Cursor soon
- OpenCode
- pi
- Claude Code
- GitHub Copilot soon
- Codex soon
- Cursor soon
- OpenCode
- pi
- Claude Code
- GitHub Copilot soon
- Codex soon
- Cursor soon
- OpenCode
- pi
- Claude Code
- GitHub Copilot soon
- Codex soon
- Cursor soon
- OpenCode
- pi
Questions the ledger answers
Not a dashboard tour. The arguments you can settle with data instead of opinion.
Where does model choice pay off?
Model-level spend shows where a cheaper tier would cut cost without cutting output.
- Opus 5.5 $1,684
- Sonnet 5.5 $962
- GPT 5.3 Codex $431
- DeepSeek v4.1 Flash $118
What does the team lean on?
Skills ranked by runs, across every harness. What is load-bearing, and what nobody touches.
- ponytail 1.3K
- svelte-code-writer 962
- plannotator 611
- supabase 384
Which tools did the team adopt?
Sessions per agent, so you know what people actually run, not what the licence covers.
Set up in minutes, then it just reports
No rollout project and no per-team setup. Install it, sign in, and your usage starts showing up.
-
Install Hashtrack
Add it to the coding tools you already use. It starts recording as you work.
-
Sign in once
Authenticate to connect the extension to your account.
-
Read the ledger
Your usage shows up in the dashboard, and you can switch it off whenever you like.
What people are saying
Engineering leads and platform teams on running Hashtrack across every harness.
We finally know what our agents cost per repo. The first invoice after rolling this out was the first one nobody argued about.
Lena Fischer VP Engineering, Northwind
One dashboard across Claude Code, Copilot and OpenCode. I stopped stitching CSVs together on Friday afternoons.
Marc Brunner Staff Engineer, Kiteworks
The skill breakdown showed us where our context budget was actually going. We cut token spend by a third in two weeks.
Aisha Okafor Platform Lead, Lumen Labs
Rollout took an afternoon. Every repo now has a cost line without anyone changing their workflow.
Tomás Silva Engineering Manager, Fieldwork
The harness view settled a long internal debate in one meeting. We could finally see what people actually run.
Priya Raman CTO, Halcyon Systems
The questions that come up in procurement
What leaves your machine, who can see it, and how to switch it off.
What does Hashtrack actually send?
Token counts, cost, model and harness names, repository and branch names, and the tools each session called. That is the whole record. There is no source code, no file contents, no diffs, and no conversation transcripts.
Does it read my source code?
No. The extension records usage while you work. It does not upload files, and it has no reason to open them. The ingest schema has no field for code.
Is this a way to watch my developers?
It measures cost and adoption, not output. Spend is attributable per developer, but attribution is counts, not content: there are no prompts, no code, and no transcripts to read. The view answers what this cost, not what this person wrote.
Which harnesses work today?
Claude Code, OpenCode, and pi are supported today. Claude Code installs the hashtrack plugin from its marketplace, OpenCode takes one plugin entry, and pi installs from npm. Codex, GitHub Copilot, and Cursor are in development and marked as soon wherever they appear, so you are never sold an integration that does not exist yet.
Can I switch it off?
Yes, at any time. Revoke the token from the tokens page and that machine stops syncing immediately. Your local usage file stays yours.
Stop guessing what AI costs.
Hashtrack turns agent sessions into a cost line you can defend: per repository, per model, per team, before the invoice arrives.