HUMANS
Built for the human eyes
Read agent output without losing your place. Measure, line height, and heading hierarchy set for ease of reading on backlit displays.
Markdown, read properly
Annotate Markdown Files
Streamline collaboration with agents over Markdown files.
Or drop a .md file anywhere on this page.
Retrieval pipeline
The indexer runs on every commit and writes chunk embeddings to the vector store. Chunk size is fixed at 512 tokens, which the retrieval layer assumes downstream.
Still 512? Ingest says 256.
Queries are rewritten before search. The rewriter shares a cache with the answer model, so a cold start costs two round trips instead of one.
- Rerank top 50 candidates with the cross-encoder.
- Drop anything below 0.42 similarity.
- Return citations with byte offsets.
Offsets are stable only while the source file is unchanged.
Evaluation runs nightly against the golden set. Regressions block the deploy unless a maintainer overrides.
Still 512? Ingest says 256.
Cache sharing is out of scope for this PR — split it out.
A reading surface for documents written by agents. Markdown is where specs, plans, and agent output already live. Annotate makes it reviewable by people and returnable to the tools that wrote it.
Still 512? Ingest says 256.> L14 · "512 tokens"
> conflicts with ingest (256)
> conflicts with ingest (256)
MACHINES
Token-efficient feedback
Send back the notes and the lines they attach to, not the whole document. Fewer tokens, fewer guesses about which section you meant.
docs/architecture.mdReview notes#4124 comments synced to the PR
DEVELOPERS
GitHub as the source of truth
Pull Markdown files from a repo, review them in place, and open a pull request when a file needs to change. The file never leaves version control.
Work together
Share a document with an unlisted link, or require sign-in to open it. Notes from everyone land on the same lines.
Review a spec a teammate wrote, or a draft an agent generated, on the same surface.
Share document
Any file as Markdown
Convert PDFs, Word docs, images, and HTML to Markdown before you send them to an agent.
Same content, fewer tokens.
Conversation
Take the three highlighted lines and rewrite the chunking section to match ingest.
Updated the section to 256 tokens and noted the downstream assumption. Two of your notes are resolved.
Connect over MCP
Your agent reads the notes itself over MCP — no copying comments into a prompt.
It gets the highlight and the lines it points at.
Repository
docsarchitecture.mdretrieval-eval.mdrollout.md
Speed up your feedback
Drop any file to review
PDFs, Word docs, and HTML convert to Markdown. Renders instantly.
Selin Korustan
Staff engineer · infrastructure team
The review round that used to take a day of Slack threads now happens in the document itself.
Omer Rabin
General Manager · TLA Venture Capital
I paste the notes straight back into the agent. It stops guessing which section I meant.
Audrey Solomon
Solution Architect · Anthropic
Cycling back my thoughts to the agent became so simple and accurate. The agent stops guessing which section I meant.