Sign in Get started

The agent compounding platform

Agents that compound.

Production infrastructure for agents: routers that optimize every call, analysts that catch every failure, and full histories so your agents compound. Simple enough to use on day one.

Get started Email the founders
npm install -g @recurz/cli && recurz

macOS and Linux. Signs you in, registers this machine and starts the collector.

Observerz mission control for an example team: 9 pieces of work running and 1 waiting on you.
The Tracez session list with spend and findings per session.
Works with
  • Claude Code
  • Codex
  • pi
  • oh-my-pi
  • Prime
  • Cursor
  • Amp
  • OpenCode
  • Antigravity
  • any OpenAI or Anthropic SDK

Start with one product, keep all of them.

Each one is useful alone. Together, every run your agents make feeds the next one. Below are the five most teams start with; Dreamz, Memoryz, Sentinelz, Tapez, Designlabz and Acceleratez are live apps too, in the product menu of every app.

  1. 01KnowsWhat did my agents do?
  2. 02SensesWhat's happening around them?
  3. 03DecidesWhat should happen next?
  4. 04ActsWho fixes it?
  5. 05ProvesDid the change help?
  6. 06TrainsCan a smaller model do it?
  7. 07RoutesWhich model should do what?

Each verb hands its output to the next, and Steerz routes the next run back to the start.

Knows · Collectorz

Collectorz: Nothing your agents do goes unrecorded.

The collector reads the transcripts your agents already write, on laptops and CI runners. It never sits between an agent and its model, and every batch waits on disk until a receipt clears it.

  • One credential per machine, revocable on its own
  • Survives reboots, dropped networks and deleted files
  • Your past sessions on disk come along on day one

More about Collectorz

npm install -g @recurz/cli
recurz
Sign in to Recurz in your browser:
https://id.recurz.ai/oauth/authorize?client_id=…
Signed in as you@company.com
This Mac is registered
Collecting Claude Code, Codex and pi in the background
Full install guide
The Tracez session list: a day of sessions with tokens, spend, a sparkline of each run and its findings.

Knows · Tracez

Tracez: The full record of everything your agents did.

Every session from every agent, laptop and CI runner, in one private place. Search it, replay it, and see where the money went, turn by turn.

  • Tokens per turn, cache reads and writes, sub-agents
  • Estimated cost by session and by kind of work
  • A retro on any session for about 7¢

More about Tracez

Acts · Handz

Handz: Catch an agent going wrong, and set it right.

Guards hold a risky command before it runs. The steering hook nudges a looping agent back on course, or hands it to you. When the same mistake keeps coming back, Handz fixes the memory or instruction behind it, and you approve the change.

  • Guards on risky commands, tried in shadow first
  • Nudge, stop or take over a looping agent
  • Memory and instruction fixes you approve

More about Handz

Handz mission control: what is running, what is waiting on you and where the spend goes.

Proves · Verdictz

Verdictz: Stop shipping agent changes on vibes.

Lab-grade statistics for every change you make to your agents: an AGENTS.md, a model, a skill, a whole harness. Did it help, by how much, and how sure are we?

  • Plan frozen and sized before anything is spent
  • Paired runs, a cluster bootstrap, one look at the planned size
  • Imports and exports Harbor tasks and datasets

More about Verdictz

A Verdictz verdict: the migrations skill is kept at +7.1 points, with its 95% interval and the checks behind it.
LIVE ROUTING steerz.recurz.ai · example data claude-code6.1/s codex4.3/s ci-runner3.2/s pi1.4/s Anthropicsonnet-5-5 58% OpenAIgpt-5.6-terra 24% Fireworksyour fine-tune 13% OpenRouterfailover 5% LLM ROUTER 15/srequests 4.2 sp50 latency $412of $1,000 budget 0 dropped1 failover, 180 ms

Routes · Steerz

Steerz: Steer every model call your agents make.

Point Claude Code or any OpenAI or Anthropic SDK at one URL with your own keys. Every call is routed by your rules, capped by your budget and metered live.

  • Failover across models and providers
  • Budgets and caps you set, enforced on every call

More about Steerz

Coming soon · Adaptive routing

Steerz doesn't pick one model for the whole session. It watches the work, and in epochs, moves model and effort up the moment a step gets hard and back down when it gets easy, holding each route long enough to keep the prompt cache warm.

Coming soon · Insights++

Smart keep-alives hold provider caches open between turns, so more of your requests are served from cache. Tracez shows every hit and miss and why, plus hidden calls and latency.

Coming soon

Senses

What's happening around them?

  • Signalz

    Your metrics and analytics, one query away for agents and for Handz.

Decides

What should happen next?

  • Decisionz

    Calibrated yes, no or which-one calls, from a fast model up to a council.

Acts

Who fixes it?

  • Fixez

    Turns errors and wrong results into fix PRs.

  • Refactorz

    Pays down messy code with cleaner PRs.

  • Oncallz

    Answers the page on your machines and clouds, with evidence.

  • Squadz

    The agent swarm everyone will run in 2028, yours today. Give it a goal and a budget. It grows agents where the work pays off and stops them where it doesn't.

Proves

Did the change help?

  • Proofz

    Catches sandbox escapes, fixes them on the spot, and signs a receipt for every run.

Trains

Can a smaller model do it?

  • Graduatez

    Trains a small model of your own on your agents' calls.

  • Pocketz

    Small models that run anywhere, even on a phone.

  • Rolloutz

    Moves repeated calls to your trained model once it matches the frontier.

Keep the next run. And every one after it.

Sign in, run the collector, and your history on disk comes too.

Get started