Teaflask

Every SaaS product forces users through menus, forms and help articles to translate what they actually want into a sequence of clicks. The example shown is an expense tool: employees have to find the right screen, enter merchant, amount and date, attach a receipt, pick a project, and route it for approval. The text frames this as an 'interface tax' charged on every click, where users don't want the interface itself, they want the outcome behind it, such as a filed and approved reimbursement.

Teaflask is an AI assistant that a SaaS company embeds inside its own product, grown from scanning that product's code so it knows the real forms, fields and APIs available. It can click, fill forms, upload files, change settings, read data, schedule things and call APIs, with each scanned action individually set to Allow, Ask or Deny. It follows written playbooks (named 'skills') that chain these actions into multi-step workflows, such as matching a receipt to a project and parking on a manager's approval. It supports durable execution that resumes after crashes, restarts or multi-day approval waits, splits long jobs across background sub-agents, re-verifies its own documentation against the live codebase after each deploy, and lets the company choose the model, memory, and whether usage is billed to the company's API key, credits, or the end user's own subscription.