Solutions · Engineering

Plan the migration before you write the migration.

Your Claude Code, Codex, Copilot or OpenCode reads the repo (read-only), opens the vendor docs on the board, and lays out options, risks and phases your team marks up live.

Claude Code is your coding harness. IdeaWave is your planning harness.

Works with

When to reach for it

Any change you'd normally whiteboard with a colleague before writing the first line. Also good for RFCs, new services and getting to know an unfamiliar codebase.

  • Migrations and refactors

    A database, framework, cloud or auth move that's too big for one PR. Agree on the options, the order and the rollback before anyone builds.
  • Incidents and postmortems

    Turn the timeline export and the code path into agreed causes and follow-ups. On-call engineers fill in owners from their own time zones.
  • Build or buy

    Compare vendors and libraries from their real docs and pricing pages, not from memory. You get a decision record with the numbers and their sources.

Walkthrough: switch auth providers without downtime

Your agent runs on your machine with your login. It can read your code but can't edit files or run commands.

You, to IdeatePlan migrating our auth to the new provider without downtime.

  1. Attach the repo

    Add your repo to the board as a folder. The agent reads the actual code, read-only, and notes what it finds.
  2. It reads the vendor docs on the board

    The agent opens the migration docs and pricing pages in browser cards, highlights the lines that matter and pins notes to them.
  3. It asks everything at once

    Open decisions arrive together as question cards: cut-over, budget, timeline. Teammates answer from the board link, in any order, whenever they get to it.
  4. The plan lands, citing your files

    Phases with rollbacks, a risk table and a flowchart of the login path, laid out on the board. The team comments and edits in place, then hands it to the coding agent.

Plans that cite real files

Attach a folder and your agent reads the code itself, not a doc about it. Each phase points at the files it touches, so reviewers check the plan against the code instead of taking it on trust.
  • Folders are read-only: the agent can't edit files or run commands
  • Local files are shared only when you add them to the board
  • Teammates comment right on the plan
How your agent connects

The docs open next to the plan

Vendor docs, RFCs, changelogs and pricing pages sit on the board as real pages. Your agent highlights what matters and pins notes on, and your team browses the same pages alongside it.
  • Highlights and notes stay on the page for everyone
  • Snapshots record exactly what the agent read
  • It reads, clicks and scrolls. It never logs in or fills in forms
Browser on the board

Hand off to your agent

When the plan is agreed, Claude Code (or any MCP client) reads it from the board through a read-only endpoint and gets to work in your repo, under its own permissions.
  • The board stays the record of what was agreed
  • Personal read-only tokens you can revoke any time
  • No browser tab needs to be open
Your agent, end to end

Bring the team. Viewers are free.

Plans get better when the people who'll build them weigh in. Share the board link; nobody else installs anything to join.

  • $0

    Viewers

    Your EM, the security reviewer, the team in another time zone: anyone who only needs to read the plan. Viewers can't comment or answer questions.

  • $5

    Per editor, per month

    Anyone who answers the agent's questions, comments or edits. Boards are unlimited.

  • 3 boards

    Free to start

    Try it on a real task first. Your agent runs on the AI plan you already pay for, so we don't bill for agent work.

Fair questions

Is my code sent to your servers?

Your agent reads your code on your machine. What ends up on the board (notes, plans, any files you add) is stored on our hosted services so your team can see it.

Whatever the agent reads goes to your model provider, under your own plan, the same as when you use the agent in a terminal.

Why not just use plan mode?

Plan mode is great for one person and one task, but it scrolls away in a terminal nobody else sees. This is for the change you'd whiteboard with colleagues: the docs, the open decisions and the plan stay on a board the team can answer, edit and sign off.

Can it write the code?

No. It reads code but can't edit files or run commands. Once the plan is agreed, your coding agent reads the board through the read-only MCP endpoint and does the work in your repo.

Is it OK with my agent's terms?

We launch the official CLIs on your machine under your own login. We never see your credentials or tokens, and agent work runs on the plan you already pay for.

Can we self-host it?

No. Boards are hosted so teammates can join from a link. You can export any board, files and all, as a single .iw file and import it again later.

Does it run on Windows?

The CLI does: npx ideawave runs on Windows x64 with Node.js 22 or later. The desktop app is for macOS (Apple Silicon) and Linux, with Windows coming soon.

Big task? Start a wave.

Open a board, attach your repo and plan the change with your team.