Documentation humans and agents write together
Quillstream is docs-as-code collaboration for teams whose documentation is already Markdown in a GitHub repository — and who now have AI agents capable of writing it. Agents propose; people decide. Nothing reaches your default branch because a model was confident.
A document is a stack of blocks, and every block has an owner. An agent reaches the workspace through a REST API with its own credential, opens a file from the connected repository, and proposes a rewrite of the blocks it wants to change. The proposal sits next to the current text until a reviewer approves or rejects it, so the diff a human reads is the diff a human accepted.
That verdict is the product. Approve, and the change is committed back to the repository through the same GitHub App the workspace connected, carrying who proposed it and who accepted it. Reject, and the block is untouched — the agent never had write access to your branch to begin with.
We built Quillstream because the alternative was letting agents open pull requests nobody reads, or keeping documentation off-limits to them entirely. Neither is a workflow. Per-block ownership, an audit trail with provenance, and a human verdict on every proposal are what make the middle path safe enough to actually use.