Never Stop Working

The problem

You're deep in a session and your AI tool hits its limit.

"You've reached your usage limit. Try again in 4 hours."

Without fai, the context you built up lives inside that one conversation. You can wait, or you can open another tool and explain your project again from scratch.

How fai handles it

fai keeps your project context outside any single AI tool. It lives in your workbench, and every agent fai has detected is pointed at it. When one tool hits its limit, open another. Your context is already there.

fai doesn't switch tools for you, and it doesn't predict limits. What it removes is the re-explaining.

What switching looks like

1. You're working in Claude Code with fai running. Claude hits its limit.
2. You open Codex in another terminal, in the same project.
3. Codex's AGENTS.md already points at your FAI-PROJECT.md, FAI-PERSONAL.md, and FAI-SESSION.md. It starts a new conversation with the same project knowledge, session history, and vocabulary.
4. You keep going. fai is still watching and reads Codex's logs, so the work lands in the same session.

The real difference

Without fai	With fai
Hit a limit, wait or start over	Hit a limit, open another agent
Re-explain your project from scratch	Context is already loaded
Each tool knows only its own history	Every agent reads the same accumulated context
Switching costs more the longer you work	Switching costs the same on session 20 as session 1

It's still a new conversation. What carries over is everything fai has learned about your project, and that grows with every session.

How each agent gets your context

fai reaches agents two ways. Most get both.

MCP server. When fai starts, it runs an MCP server and adds it to the config of each detected agent that supports MCP. Those agents get live access to fai's tools and context.

Direction files. fai also writes a marked <!-- fai:context --> block into each agent's own rules file, pointing at your three context files:

Agent	Direction file
Claude Code	CLAUDE.md
Codex CLI	AGENTS.md
Cursor	.cursor/rules/fai-context.mdc
Gemini CLI	GEMINI.md

The block is rewritten every time you run fai. The files it points to keep updating as fai learns. Agents without MCP support, like Aider and Amazon Q, get their context this way. See Agents for every agent's file.

This is what makes working across agents possible: whichever one you open, it reads the same context.

Next: Right AI for the Job - Configure the right model for each task