turnset Talk to us

Real human sessions with agents, licensed with the rights attached.

Turnset supply is not scraped from public text or harvested out of product telemetry. Every trace is a person directing an AI agent through work they actually needed done, contributed under an explicit permission grant, with the context and the result intact.

What's different

Genuine interaction, not a log dump.

The left column is what most workflow data on offer actually is. The right column is what a Turnset trace carries.

Logs and telemetry, harvested from a product's users

Sessions contributed on purposeThe contributor starts the session deliberately and has a window to remove anything before it is licensed. Nothing arrives by default.

Final answers, lifted from public text

The whole exchange that produced the answerPrompts, replies, tool calls, file reads, edits and diffs, in the order they happened.

Clean one-shot examples that resolve on the first try

The failures and the recovery after themWhat broke, what the error said, and how the person redirected the agent to get out of it. This is the part that cannot be synthesized.

Text with no evidence that the work succeeded

Outcome signals from the session itselfTest runs, exit codes, retries, and whether the task actually resolved.

Unclear origin and an unclear rights chain

A provenance record per traceHow it was contributed, under which terms version, when it was captured, when it finalized, and what was removed before packaging.

Why the human half matters. Model output is abundant and getting cheaper. What stays scarce is a competent person deciding what to ask for next, judging a wrong answer, and steering an agent through a task that did not go straight. That decision sequence is what a trace preserves.

Provenance

  • Consent is recorded per session, not assumed once. Contributors accept the terms, then choose which sessions to capture. Each trace carries the terms version in force when it finalized.
  • Rights are asserted at capture, not inferred later. Personal, open-source and sandbox work only. Employer code, client code and anything under an NDA are out of scope, and that boundary is stated to contributors before they start.
  • Removal runs forward, never backward. Every trace is visible to its contributor for 24 hours and can be excluded free in that window, so what reaches a package has already passed it. A contributor can also withdraw a trace later, in which case we drop it from every future package and reverse the payment we made for it — but copies already delivered under a license stay licensed. Nothing you have bought gets recalled.
  • Secrets are removed on the contributor's machine. Tokens, keys and identifying paths are scrubbed at capture rather than cleaned up afterwards. We describe the result as privacy-preserved, not fully anonymized, because traces this rich do not support the stronger claim.

Tell us what you need it to cover.

Supply is being built now, contributor cohort by contributor cohort, which means the shape of what we collect is still open to influence. If you have a specific gap — a language, a class of agent, a kind of task, a rights posture your legal team requires — that is worth a conversation early rather than late.

Talk to us