Describe the job
One plain sentence: what the agent does and who it serves. The builder drafts instructions and starters on your own key.
Product · Build
Assistants that know their job.
Custom AI agents with instructions, starter prompts, and knowledge files. Describe the role and the builder drafts it.
The refund policy stopped being tribal knowledge the day the support agent actually read it.
- AI drafts the agent from one description
- 35 templates across 8 role packs
- Up to 10 knowledge files per agent
- Starter prompts for the first message
1 sentence ina working agent out, with instructions and starter prompts you can edit
How it works
Attach knowledge
Policy PDFs, tone guides, product sheets. Their text grounds every reply the agent gives.
Chat or mention
Open an agent chat, or type @ in any conversation to hand the thread to it.
01
From idea to working agent in minutes.
You do not write a system prompt. You write a job description, and the builder turns it into instructions and starter prompts you can edit like text.
- Plain-language agent builder
- Improve-instructions pass on demand
- A default model per agent, still switchable
02
Or install a pack and skip the blank page.
Five role packs cover the common desks: founder ops, marketing, support, sales, and product. One click installs the pack with sensible model choices.
- 17 ready-made agent templates
- Packs mapped to your cheapest capable models
- Every template editable after install
Questions, answered
No. Describe the job in one plain sentence and the builder drafts instructions and starter prompts, all editable as text.
Up to 10 knowledge files per agent. Their extracted text grounds every reply the agent gives.
Each agent has a default model, and you can still switch it in the composer for any single chat.
Size it with a free tool
Browse reusable prompt templates
Start from a structure that already works, then adapt it.
Validate a tool definition
Catch schema errors before an API call rejects them.
Why startups need a model-agnostic AI workspace
The case for keeping model choice open as your team grows.