Cycles

Appearance

Design Partners ​

Cycles is in design-partner mode. We partner with a small, focused set of teams running real agent workloads — multi-tenant, concurrent, with real cost and real consequences when something goes wrong — who want hard runtime limits on spend, risk, and agent actions, and who are willing to push us hard on what's missing.

This is not a generic beta. The goal is to put Cycles under real workload pressure, find what breaks, and shape the next two protocol releases around production use.

Built by Albert Mavashev (ex-CTO Nastel, ex-meshIQ) and a small team with nearly three decades building mission-critical transaction infrastructure for banks, airlines, and telecoms. Cycles applies that same discipline to AI agent execution.

What design partners get ​

  • Direct line to the founder. Monthly 30-minute office hours. Email or Slack any time in between.
  • Roadmap influence. What ships in the next two minor versions is shaped by the workloads design partners run. Your incident becomes our next runbook.
  • Free self-hosted forever. Because Cycles is Apache 2.0, your self-hosted deployment remains free to run. You run it in your VPC; nothing leaves.
  • Priority support during your 60-day integration window. Private Slack/email channel and fast response for production-blocking issues.
  • Optional public co-marketing. Joint blog post, case study, or conference talk — opt-in, not required. You can be a named partner or stay anonymous; your choice.

What we're asking for ​

  • One real workload running on Cycles within 60 days. Not a sandbox, not a Hello World — something with real traffic, real concurrency, and an operational consequence if it fails. Internal tools and dev pipelines count; tutorials and synthetic load tests do not.
  • Monthly 30-minute sync. Office hours go both ways — we need to hear what you're hitting.
  • Honest feedback. We want pushback, not validation. If the protocol is wrong, the comparison page is misleading, or the SDK is awkward — we need to know.
  • Permission to use anonymized learnings unless otherwise agreed. Named case studies and public co-marketing are opt-in.

Who this is for ​

  • Teams running multi-tenant or concurrent agents with real workloads — internal tools, dev/research pipelines, customer-facing systems, coding agents, ops automation. Anything where real cost is being burned and operational risk is no longer theoretical. Already running, or shipping within 60 days.
  • Teams running Python, TypeScript, Spring Boot, or MCP-host stacks (Cursor, Continue, Windsurf, etc.). Rust support is welcome if you're willing to help shape it.
  • Teams that can move fast. No 6-month procurement cycles. We need to ship, learn, ship again.
  • Teams whose engineering leadership is comfortable with self-hosted Apache 2.0 OSS. There is no managed cloud yet.

Who this is NOT for ​

We'd rather be honest now than waste both of our time later. Cycles is probably not the right fit if:

  • You're running a single-tenant prototype with no concurrency. You don't need atomic reservations yet — a local counter is fine.
  • You're still in exploration mode and don't expect to put Cycles in front of a real workload within 60 days.
  • You need a fully managed cloud / SaaS. We're self-hosted-only today; managed cloud is on the roadmap, not live.
  • You can't commit to running one workload on Cycles within 60 days. We're not in a position to help with multi-quarter evaluations.
  • Your blocker is whether agents need governance at all. The design-partner conversation assumes that's already settled for you.

Apply to become a design partner ​

The fastest path is to grab 30 minutes:

Book a 30-minute call →

Or email [email protected] with one paragraph on:

  1. What workload you'd run on Cycles
  2. What stack you're on (language, framework, agent host)
  3. What's currently broken or worrying you about agent budgeting / actions

I read every email; expect a reply within 48 hours.

What happens after the first call ​

  1. Mutual fit conversation. No commitment yet. We talk through the workload, the stack, and whether Cycles actually solves your problem.
  2. One-page MOU. If fit looks right, we draft a one-page commitment doc covering scope, timeline, and what each side delivers.
  3. 60-day integration window. Hands-on support, weekly check-ins as needed.
  4. Public artifact at 90 days. If both parties want it: case study, joint blog post, or conference talk. Opt-in.

Why we run this program ​

Runtime authority for AI agents is a category that did not exist 18 months ago. We can write about it, comparison-page it, and demo it — but the product gets built right only if real teams put it under real load and tell us where it breaks. A small, focused cohort is enough to drive every meaningful product decision for the next two quarters. We'd rather get those right than pursue scale prematurely.