Permissioned AI workflow data
Get paid for the work you're already doing with AI.
You run agents all day and every bit of that value stays with the platforms. Turnset captures the sessions you start, shows you every trace on the web, and pays you for each one that's kept.
This phase: GitHub or Google sign-in · macOS · terminal agents · personal, OSS & sandbox work · paid participation opens in the US first
- Agent
- claude code · sonnet
- Captured
- 47 min · 61 tool calls · 4 files
- Work
- bug fix in a personal repo
- Rights
- yours · personal project
- Scrubbed
- 2 secrets, 1 path
- Your turns
- 23
- Your tokens
- 8,412
- Quality
- ✓ verified outcome · tests pass
- Window
- closed - 24h after capture
Illustrative example - not real contributor data
Scope
What leaves your machine, and what never does.
Turnset runs only for the session you explicitly start. It is not a background daemon, and it is not watching the rest of your computer.
Captured in a started session
- Your prompts and the agent's responses
- Tool calls: edits, shell, search, file reads
- Diffs and the files the session touched
- Outcome signals: tests, exit codes, retries
- Model, agent, and timing metadata
Never captured
- Anything outside the session you started
- Keystrokes, screen, clipboard, browser
- Other terminals, other apps, other repos
- Secrets and tokens - scrubbed before packaging
- Your identity in any licensed dataset
Contribute personal, open-source, or sandbox work only. Not employer or client code, and nothing under an NDA. The rights have to be yours to give, or we can't license it and you shouldn't offer it. This is the tightest rule in the pilot and it's deliberate.
Payment
A bonus when you join, then paid per trace.
You are not waiting on a sale and you are not on a royalty. Each trace is priced on its own once it finalizes. Three things set the rate.
| Rate basis | What it measures | Why it matters to buyers |
|---|---|---|
| Your turns | How many turns in the trace are yours | Multi-turn direction is the scarce part - it can't be synthesized |
| Your tokens | How much you actually wrote across those turns | Substance, not one-word nudges |
| Trace quality | Tool use, verified outcomes, recovery from failure | Sessions that resolve something are worth more than ones that wander |
Sign-up bonus first. You're paid a bonus for joining the pilot, before you've contributed anything, so you're never working on spec. After that each finalized trace is priced on its own.
Being straight about the fine print. We set how payment is calculated and can change it at any time - a change applies to traces captured after we publish it, never to work you've already done. Every trace runs through fraud, duplicate, and rights checks before it pays, and whether a trace is paid, how much, and when are our determination, at our sole discretion. A trace that finalizes is licensed to us whether or not we pay for it. Read the Terms, sections 11 to 13, before you sign up - that is where the binding version lives.
Setup
Two commands, then your normal workflow.
Install
One binary. macOS in the first build.
curl -fsSL https://turnset.ai/install | sh
Start a session
Your agent, launched through Turnset.
turnset start <claude|codex|cursor|opencode|openclaw|hermes>
Work as usual
Same agent, same models, same habits. Traces from the session show up in your dashboard on the web.
turnset.ai/traces → 3 new traces · 24h to exclude
There is nothing to install yet. This is what setup will look like. The first macOS build goes to the earliest phase, and the list is how you get into it.
There is nothing to finalize by hand. Every trace sits in your dashboard for 24 hours. Exclude any of them in that window and they're gone; leave them alone and they finalize on their own and pay out. Checking in is optional, not homework.
Pilot boundaries
- Region
- United States
- Platform
- macOS
- Environment
- Terminal agents
- In scope
- Personal, OSS, sandbox
- Payment
- Per trace + join bonus
- Exclude window
- 24h, then auto-finalize
Working in an app like Claude or ChatGPT, or in VS Code or Cursor? Join anyway - those environments come after this phase, and the list is how we decide the order.
Demand
Why anyone pays for this.
AI performance is limited more by data quality than model size, and organizations now care as much about provenance, rights, and governance as they do about volume. Permissioned datasets with documented lineage are scarce. Turnset packages and licenses them, and that revenue is what funds contributor payments.
- Model development
- Evaluation
- Reinforcement learning
- Agent benchmarking
- Product research
- Academic research
- Enterprise AI
Questions
The things people actually ask.
- Who is this for?
- Developers and power users who work with AI every day. The first cohort covers terminal agents: Claude Code, Codex, OpenCode, OpenClaw, Hermes. Apps and editors come next.
- Can I change my mind?
- Every trace is visible in your dashboard for 24 hours after it's captured. Exclude any of them in that window and they're gone. Once a trace finalizes it can be licensed, and licensed data can't be retrieved - which is why the window exists.
- Do I have to review anything?
- No. Excluding a trace is optional. If you never open the dashboard, traces finalize on their own after 24 hours and pay out.
- Why does paid access open in the US first?
- Paying people for data carries different obligations in every jurisdiction. We'd rather open one region properly than all of them badly. Join from anywhere - the list is global, and we'll email when your phase opens.
- Is my data anonymized?
- It's privacy-preserved: secrets scrubbed, identifiers removed, packaging designed to avoid re-identification. We don't claim "fully anonymized" because no one honestly can.
- Do I have to change how I work?
- No. Turnset is vendor neutral and runs alongside the agents, models, and tools you already use.
- How do I get in?
- Join the list with your email, or sign in with GitHub or Google. Either works. Signing in verifies your address and puts you in this phase's preferred group, alongside macOS and terminal agents.
- Why does signing in help?
- It tells us the address is real and gives us a handle to recognise you by, which is most of what we need to open access in a sensible order. We read nothing else - not your repositories, not your profile.
Own the value of your AI workflows.
One list, opened in phases. Email is enough; signing in with GitHub or Google puts you further up this one.