Cycles — Runtime authority for autonomous agents

2 min read Original article ↗

CyclesRuntime authority for autonomous agents

Stop runaway agent spend and risky actions before they execute.
Self-hosted, no prompt storage. Apache 2.0.

Cycles hero image

Cycles hero image

See it in 60 seconds

Cost runaway

Same agent. Same bug.

Without Cycles: ~$10 in 12s — the pace behind $4,200 incidents.

With Cycles: $1 cap, before the next action ran.

reserve → deny → no downstream call

Run the runaway demo →

Blast radius

Same workflow. Different boundary.

Without Cycles: All four actions execute — including the customer email.

With Cycles: Internal actions proceed; the email is blocked before it sends.

decide → DENY → no email send

Run the action-authority demo →

Add runtime authority in a few lines

@cycles reserves budget before the action runs. No remaining cycles — no action.

from runcycles import cycles

@cycles(estimate=5000, action_kind="llm.completion", action_name="openai:gpt-4o")
def ask(prompt: str) -> str:
    return openai.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": prompt}]
    ).choices[0].message.content
How-To icon

How-To Guides

27 integrations: OpenAI, Anthropic, Groq, AutoGen, Spring AI, LangChain, LangGraph, AnyAgent, Vercel AI, Next.js, Django, Flask, MCP, and more. Budget patterns, estimation, error handling.

Browse guides

Protocol icon

Protocol Reference

Reserve-commit lifecycle, authentication, scope derivation, units, caps, overage policies, TTL, decide, dry run, events, debt, balances, metrics, and error handling.

Read reference

From demo to enforcement, in 5 steps

Most teams reach out for one of four reasons — they all land in the same inbox, and a founder reads every one.

  • Piloting Cycles on a real stackYou're putting Cycles in front of a real workload — Python, TypeScript, Spring Boot, Rust, MCP — and want implementation help or a second pair of eyes on the integration.
  • Hit a failure mode we haven't documentedRunaway agent, retry storm, unit mismatch, scope-derivation edge case. Tell us what broke; we'll add it to the runbook.
  • Feedback on the protocol or roadmapYou've read the spec and have opinions on the error contract, permission model, or something you need in v0.1.26.
  • Evaluating Cycles for productionYou're evaluating Cycles for production — comparing it to rate limiters, LLM gateways, or observability tools. We'll walk through the spec, demo the runtime, and answer integration questions.

Running a real agent workload and evaluating Cycles? Apply to become a design partner →

Found a bug or have a concrete technical issue? Open a GitHub issue instead — it’s faster for everyone.