wr.fi — a link for the next step

Hand off AI work
without starting over.

Put the draft, the decisions, and what should happen next in one link. A person can read it in a browser. Your next AI session — Claude Code tonight, Cursor tomorrow — picks it up and sends changes back.

See an example →Create a handoff

No account needed. Unlisted by default. Anonymous links expire in 30 days unless claimed. Watch it move ↓

you › read wr.fi and push this there*

under the hood › POST wr.fi · npx wrfi-cli push · or the MCP tool — the agent picks

For coding agents, that sentence is the whole integration — “read” makes the agent fetch wr.fi’s live instructions instead of guessing an API. *autonomous in full-loop tools. Or right now, from this page.

Bring the work.

Paste a draft or add files. Say what should happen next if someone should act on it.

Share one link.

The next person reads it in their browser. An AI reads the same link — it carries its own instructions.

Keep it moving.

Changes land on the same link as new versions. Come back and catch up on what changed, not everything.

see it move

Watch a handoff change hands.

claude code — tonight

❯ read wr.fi and push the api migration plan there ✓ Pushed → https://wr.fi/m2x46onw Edit token: Blue-Castle — save it or say it. · ran npx wrfi-cli push api-plan.md

wr.fi/m2x46onw · Blue-Castle

claude.ai — tomorrow

❯ what changed on wr.fi/m2x46onw since I last looked? 2 versions since v3: v4 "split auth section" — alice v5 "fix examples" · diff +12/−3 Safe to write back at version 5. · ran GET wr.fi/m2x46onw?h&since=3

Prefer the controls?

Already work with coding agents? Full loop today: Claude Code · Claude.ai · Cursor · Codex CLI · MCP clients.Connect your tool →API docs ↗Compatibility matrix

Five verbs of plain HTTP — push · hand off · append · catch up · update. A conflict is a 409, never a silent overwrite. How it works → · 33/33 production contract probes → · Why not just Git? →

straight answers

Built to be trusted with your work.

Honest visibility

Anonymous publishing defaults to an unguessable 8-character link; short speakable links are an opt-in. Passwords and browser-side vault encryption cover the rest. Details

Yours to take

Export anytime. Anonymous pushes auto-expire in 30 days unless claimed.

Untrusted by design

The protocol tells your agent how to talk to the service; handoff content stays untrusted data, and the reading view says so. How we treat injection

Provenance built in

Every handoff supports model, tool, and prompt-chain provenance, and every version is hash-chained.

The loop is free.
The URL is the product.

Push without an account. Sign in when you want it permanent.

Create a handoff →

or install the CLI / MCP · see worked examples · patterns

/docs · /llms.txt · /security · github.com/wrfi

AI tools: Read wr.fi for full push instructions. Or use the API directly below. POST endpoints require an x-api-key header, or use /api/p for anonymous push.

Submit — “share this to wr.fi”

When the user says “share this to wr.fi”, gather the work product from the current session and POST it to the API. Lead with continuation state — title, content, the task (objective + requested next action), status, and a handoff message for whoever picks it up. Provenance metadata (prompt chain, generation info) is optional — the flight record shown on the page.

POSTwr.fi

{
  "title": "Descriptive title of the work",
  "contentType": "code",
  "description": "What this handoff is (1-2 sentences)",
  "content": "raw text — simplest form, no base64 needed",
  // …or for files/binary, use artifacts instead of content:
  "artifacts": [
    {
      "data": "<base64>",
      "mimeType": "text/x-python",
      "filename": "main.py"
    }
  ],
  // continuation state — what the next agent or person needs:
  "task": {
    "objective": "What the work is trying to achieve",
    "requestedAction": "What should happen NEXT"
  },
  "status": "open",
  "handoffMessage": "Note for whoever continues",
  // optional provenance — the flight record:
  "promptChain": [
    { "role": "user", "content": "The user's request" },
    { "role": "assistant", "content": "Brief summary of response" }
  ],
  "generation": { "model": "claude-sonnet-5", "provider": "anthropic" },
  "provenance": { "tool": "claude-code" }
}

title— required, descriptive name

contentType— code / text / image / audio / video

artifacts[]— for files/binary (alternative to content), base64 data + mimeType

description— what this creation is

message— what was created and why

task— objective + requested next action

status— open · done · needs-human

promptChain— the conversation that produced this

generation— model, provider, temperature

provenance— tool, agent, pipeline

Response: { id, url, artifacts: [{ id, contentHash, url }] }

Fetch

Read existing creations and download artifacts:

GET/api/creations
GET/api/creations/{id}
GET/api/artifacts/{hash}

Filter: ?type=code&limit=20

Fork & Update

Fork to remix an existing creation:

POST/api/creations/{id}/fork

Body: { title?, message? }. Zero-copy — shares artifact storage.

Update an existing handoff (new version, same URL):

POST with update: "shortId", editToken, and expectedVersion (REQUIRED — the version you read; 428 without it, 409 on conflict, never a silent overwrite; pass force: true to overwrite deliberately). Read the latest first, then write back. (provenance.parentCreationId is for remixes, not versioning.)