Open source · AGPL-3.0 · self-hostable

Paste a link.
Review a spec.
Zero signup.

Specs live in GitHub and Confluence, but their reviews die in PR threads and meeting chat. Kedge turns any source into one readable review page: discussion anchored to the text, approvals pinned to versions, and AI on tap to strengthen the doc.

Public GitHub & raw URLsMarkdownMDXHTMLPlantUML & Mermaid via Kroki

No URL handy?

Comments that keep their place

A real review in Kedge: the doc rendered for reading, threads pinned to the exact sentences they question, approvals pinned to the version they saw. Ship a new version and the discussion re-anchors instead of washing away.

kedge.review
A spec rendered in Kedge, with comment threads anchored to sentences in the text and version-pinned approve and re-sync actions

Review where the doc is readable, not where it’s stored

Three steps from a URL to a running review. No migration, no new home for your docs, and reviewers never need accounts.

  1. Import from anywhere

    A GitHub file, a whole repo directory, a raw URL, an upload. Diagrams render, broken markup never crashes the page, and a share link exists in seconds.

  2. Discuss on the text itself

    Select a sentence, start a thread. Propose the fix as a suggested edit the author accepts in one click. Reviewers join by magic link, no accounts to herd.

  3. Ship a new version, keep the review

    Re-sync pulls the latest doc and every comment finds its place again: exact match first, fuzzy match second, the orphan tray as the safety net. Approvals show exactly which version they vouch for.

In development

Your next reviewer might not be human

Kedge’s MCP server makes AI agents first-class reviewers. They read the rendered doc, anchor comments to real passages, and wear a badge that says exactly what they are. Nothing an agent writes posts without a human confirming it.

Then close the loop: fold the review into an AI digest, hand your coding agent the improve-prompt, re-sync the revised doc, and the whole discussion is still there.

Projects & tracked repos

Point it at your docs/ folder and every spec arrives review-ready

A project tracks a repo, a branch, and a path pattern. Kedge previews the matches, bulk-imports them, and a re-scan after each push imports new files, re-syncs changed ones, and flags deletions. It never deletes anything itself.

  • Live import status on every doc, retry inline
  • Lifecycle, open threads, and sync state at a glance
  • An Unfiled bucket, so nothing needs a home first
kedge.review/workspace
Kedge v1Active
⎇ kedgehq/kedge · main · docs/**/*.md
6 docs11 open1 orphanRe-scan
Platform RFCs3 docs
⎇ kedgehq/platform · main · rfcs/*.md
Unfiled2 docs
drag docs here… or anywhere
Document
SPEC.mdIn reviewv4 · 7 open · 2 h ago
DESIGN.mdApprovedv2 · 0 open · 2 h ago
rfc-017-anchoring.mdx1 orphanv6 · 3 open · 40 min ago
specs/m4-ai-agents.mdImporting…found by re-scan · just now
ROADMAP.mdApprovedv3 · 1 open · 2 h ago
legacy-import.htmlFailedRetryunreachable · 1 d ago

Live today, charted next

Six releases in July, each one demoed on Kedge’s own docs the week it landed. The chart from here:

Live today

  • Import from GitHub, raw URLs, uploads, private repos
  • Anchored threads and one-click suggested edits
  • Versions, diffs, and approvals that go visibly stale
  • Workspace projects and tracked repos with re-scan

Charted next

  • AI reviewers and digests over MCP, in development now
  • A review inbox: mentions, digests, what needs you
  • GitHub App and Confluence sync: push once, it re-syncs itself
  • Self-host GA: one command, full parity, AGPL-3.0

Both editions launch together: the SaaS for anyone with a URL, the compose file for teams whose specs stay home.

Your specs never have to leave home

The open-source edition is the entire product: anchoring, approvals, the MCP server, all of it. AGPL-3.0, full parity with the SaaS, one compose file.

Start your first review in 60 seconds

Leave your email to open a workspace for your team’s specs. Nothing to install, no credit card, and the demo works before you decide.

Want proof first?