AgentGraft

Documentation

AgentGraft puts an AI assistant inside your Django app. You describe its tools, agents and voice in one Python file, push it with one command, and show it with one template tag. Your code stays the source of truth; AgentGraft runs the conversation, the streaming and the memory.

Pages

For agents

The docs are also published in the llms.txt format, so an AI assistant can read all of it in one fetch. Plain text, no auth.

fetch("https://agentgraft.com/docs/llms-full.txt")

How it's built

Configuration lives in code

Every tool, agent and prompt is in one Python file in your repo. The dashboard is read-only, so nothing drifts away from the code.

Sync is all or nothing

One command pushes the whole file, and the server applies it in a single transaction. To roll back, revert the commit and sync again.

Identity stays in your app

AgentGraft never stores your users. The widget signs an opaque id, and your @ag.user_loader turns it back into your own user inside your tools.