rethinker
private beta

Claude coding agents, made practical for real engineering work.

rethinker builds tools for teams that use Claude Code on real codebases. Our first product, now in private beta, keeps long agent sessions focused: short handoff notes, clearer prompts and in-the-moment guardrails. Next: slice-by-slice legacy modernization, run by Claude agents and reviewed by engineers.

Built for Claude Code · also works with Codex CLI and GitHub Copilot

Illustration of the messages rethinker shows, not captured output.

What we do

Our first product is in private beta with a small number of external teams. Legacy modernization is what we are building next.

01 · now · private beta

Focused coding-agent sessions

In long agent sessions, a large share of the cost is re-reading context, not writing code.

  • Smart handoff: a short note replaces a bloated transcript, and a fresh session picks up from it
  • Guardrails: oversized context, a cold prompt cache and runaway loops, flagged in the moment
  • Prompt refinement: clearer asks, fewer wasted turns
  • Cost view: what each task cost, computed from Claude's usage data, cache writes and reads included
02 · next · in development

AI-native legacy modernization

We are designing a way to rewrite old systems slice by slice instead of in one risky big bang.

  • An inventory graph of the legacy app: screens, endpoints, data, dependencies
  • Small slices, each with acceptance criteria and a rollback plan
  • Claude agents will do the migration work; engineers review and own every change
  • One progress metric everyone can read: how much has really moved

Built on Claude

rethinker exists because Claude Code is good enough for real engineering work. We make long sessions with it sustainable.

claude code

In the session

rethinker works inside Claude Code sessions: it watches context size, cache state and repeated commands, and suggests a handoff at the right moment.

claude models

Handoff notes

Handoff notes are written by Claude at the end of a session, so the next session starts from a summary Claude itself can use.

prompt caching

Cache-aware by design

Suggestions are timed around prompt-cache behavior, so stable prefixes stay cacheable instead of being broken.

Principles

What you can expect from us.

  1. 01
    Measured in money, not token counts.Shorter prompts are not always cheaper: cache writes and reads change the math. When we report savings, we report them in money, including our own overhead, or say plainly that nothing was saved.
  2. 02
    Your code stays yours.We will document exactly what rethinker reads, what it sends to the Claude API and under whose key, before general availability.
  3. 03
    Small, reversible steps.Every modernization slice will come with acceptance criteria and a rollback plan. No big-bang rewrites.
  4. 04
    Teams, not leaderboards.We help engineers work better. We never rank individuals by usage.