The agent only knows what you show it.
Every AI tool says it uses your context. Abstract shows you the context: which files, which folders, which knowledge, visible in a panel and changeable per document.
Context is a decision, not a black box.
The difference between a generic draft and a usable one is what the model knew when it wrote. Abstract makes that an explicit, inspectable decision you control, instead of a retrieval system you have to trust blind.
A knowledge base in plain files.
Company vision, KPIs, vocabulary, customer lists, tone rules: a folder of Markdown files the agents can draw on. Always-on for some projects, invoked on demand for others. Because it is files, it is searchable, versionable and yours.
Project instructions.
Each project can carry its own standing instructions: audience, format rules, what never to touch. Agents read them before every run.
Sources in, work out.
Meeting transcripts from Granola, Fathom or Fireflies land in a project folder and become context. Connectors bring in tickets and records from the tools your team runs on, and push updates back out. Transcript and PM-tool connectors are on the way; today, anything you can save as a file already works.
Local-first, and not a hostage.
Your documents live on your machine as ordinary files. Point other tools at the same folder if you like. Export or publish when you want to share; delete Abstract and your work is all still there.