How the trace market works
Labs have teams working on different problems, a bit like quants. They acquire data, turn it into tasks, and test whether it actually improves a model.
That data comes in two forms. Company operational data: Slack, email, code, CRM. And agent traces, from consumer conversations, people using coding agents, or automated agents. We help with both.
Agent traces can come from consumer conversations, people using coding agents, or automated agents. Traces are valuable to labs because they can be used to find agent failures and help fix them.
Traces show labs where agents fail, and help fix them. They also carry context about the world, through web search and integrations, and about the person. In many ways, someone chatting with an agent is distilling their expertise.
Coding, healthcare and finance interest us most. Your sector decides which buyers your traces are valuable to.
Who we are
We started out deploying OpenClaw to enterprises, and with a small OpenClaw SaaS. But we were quickly approached to sell our OpenClaw traces.
Because we sold to enterprises, we had no interest in selling their data, but instead helped other startups sell their agent traces.
How we work
We qualify your traces and data, then share the right opportunities with labs from our internal marketplace.
We don't buy data. We sell it to labs and other data companies, usually in non-exclusive deals, so the same data can sell more than once.
We take 20% of each deal we bring in. That may change as we learn what it costs to operate at scale, but we want early partners to keep as much of the upside as possible.
What makes data valuable
These are what we'll ask about. Rough estimates are fine.
- Sector
- Finance, medical, coding and other workflows. Niche users, like designers or doctors, can be especially valuable.
- Source
- Customers using your app, your team using coding agents, or automated runs. Keep them separate.
- Volume
- Traces or full sessions, the dates they cover, storage size, and how much arrives each month.
- Models
- Which models and versions, and how the data splits between them.
- Users
- Monthly active users, where they are, and how they're distributed.
- Harness and tools
- Custom, Codex or Claude Code; whether it has a sandbox; the tools and integrations users connect.
- Session length
- Deep research and long coding runs are the interesting ones. Exact numbers aren't needed.
- Outcomes
- Whether the agent succeeded: user corrections, ratings, tests, merged PRs.
- Terms and permissions
- What your terms let you license, including past versions, the dates they changed, and the data under each.
- Existing deals
- Anything you've already licensed, and any exclusivity.
Personal traces
If you don't have an app, or your volume is low, your personal Codex or Claude Code traces have been hard to sell on their own.
Sell Traces and Sell Codebase let you submit private traces and codebases for licensing, and earn when a deal happens. Private traces take more work, so we pay out 70% of what we license them for.
Both are in early alpha and open to US users; everyone else can join the waitlist. To license your whole team's traces, book a call instead.
- Sell Traces selltraces.com
- Sell Codebase sellcodebase.com
- Book a call
Next steps
Start with a description and rough stats, not raw data. Please don't send private records, credentials or unredacted logs.
Our questions are long, so give the prompt below to your agent and it will write the summary for you. If there's a fit, we'll ask for a small approved sample, and work through permissions and terms with you before anything goes to a buyer.
We can't promise an offer or timeline before we've seen the data, but the more detail you give us, the faster we can tell you what makes sense.
- Take the survey through your AI
- Get in touch Book a call