# Templates - Obss

Run experiments
Templates
Start an experiment from a Home template, such as Getting started, Stripe interface evals, Latest model shootout, or Skill lift, then review and launch it.
A template adds ready-made tasks, treatments, tools, and a machine to your project, so you can launch your first comparison in minutes. To start from scratch instead, select
New experiment
in the sidebar.
Start from a template
1
Open Home
The
Templates
section on your project’s
Home
page lists each template with its task and treatment counts. After your project runs its first experiment, the section starts collapsed. Select
Templates
to open it.
2
Select a template
Select a card. Obss adds the template’s tasks, treatments, and tools to your project and opens
New experiment
.
3
Review the experiment
Check the tasks, agents, treatments, and trials, and
change any of them
.
4
Launch
Select
Launch
.
A card marked
Needs a key
has a tool or machine that needs a value, such as an API key, before a run can use it. After you select it, follow the notice on
Home
to
add the value
.
Available templates
Template
What it sets up
Getting started
A short starter set to see an agent inspect, fix, and document code
Firebase vs Supabase
Compares the Firebase and Supabase client SDKs on the same create, query, and update tasks
Stripe interface evals
Compares a Stripe skill, MCP server, CLI, and SDK, with every task in Stripe test mode
Latest model shootout
Runs the same varied tasks across every available agent, to show which does your work best and at what cost
Skill lift
Shows whether a skill helps agents on React tasks: no skill, then skill version 1 and version 2
Redis on a custom machine
Uses a custom machine with Redis running next to the agent. The agent stores and reads realistic data in it.
Firebase vs Supabase
and
Stripe interface evals
need your own keys for those services.
Edit what a template creates
A template creates ordinary tasks, treatments, and tools. You can edit them, launch them again, and use them in other experiments like anything you built yourself. They follow the same rules: files and machines attach to tasks, tools attach to treatments, and you choose agents at launch (
Where files, machines, and tools attach
).
Next steps
Launch an experiment
Change trials, agents, and Evals before you launch.
Read results
Compare Evals results and cost in Matrix.
Tools
Build your own skills, MCP servers, CLIs, and SDKs.
Agents
See the agents a model shootout can compare.
Launch an experiment
Schedules
Powered by
This documentation is built and hosted on Mintlify, a developer documentation platform