Synth Workshop · Onboarding
First success is a completed research loop.
Increase your Experiments Per Minute Activation means finishing useful work—not merely downloading, signing up, or opening the app.
Use Workshop locally
Download the macOS app and begin without an account. Local models, local data, and local execution are enough to start a research loop.
- 1. Download the macOS app and verify the checksum on the download page.
- 2. Launch Workshop. Choose a local model or ChatGPT login—no Synth account required.
- 3. Run one eval or environment QA pass and inspect the evidence.
- 4. Change one thing and run again. That is the first Experiments Per Minute loop.
Use Workshop with Synth Cloud
Create an account, receive included usage, connect the app, and unlock account-backed eval, optimization, SFT, and cloud-model work.
- 1. Create a Workshop account and confirm included usage is visible.
- 2. Pair the macOS app to the account from the device flow.
- 3. Start an account-backed eval, GEPA, or SFT job with an explicit budget.
- 4. Inspect live progress, then the terminal result, then the next experiment.
Pick a first workflow
Evals
Run and compare evaluations with evidence you can inspect, not a score you have to trust blindly.
Environment QA
Build, exercise, and debug environments until their behavior is stable enough to optimize against.
GEPA and optimizers
Improve systems with evaluation-driven search. Workshop consumes Optimizers as an independent service, not a hidden internal.
Supervised fine-tuning
Train, checkpoint, and evaluate with the same evidence loop as evals and GEPA.
Connect a provider when you need one
ChatGPT login
Use the ChatGPT account you already work with. Workshop does not require you to abandon that path.
Local models
Run on-device models without an account. Local inference stays on your Mac.
Direct provider APIs
Bring your own credentials. Provider keys remain in the native app's custody.
Synth API
Account-backed Synth services add managed capability, shared usage, and cloud scale without changing the research loop.