zocrity 0.1.0

Budget-Safe AI Agents with Local Caching and Circuit Breakers.

🛑 Zocrity — The Emergency Brake for AI Agents

Stop API hallucination loops before they drain your budget. Zocrity wraps your agent runs with a hard stop so runaway retries, repeated prompts, and token spirals don’t turn into surprise cloud bills.

---

🚀 For Replit Users (The 1-Click Install)

If you’re already in this workspace, just hit Fork and you’re basically done.

Then feed your own Agent this exact prompt:

Agent: Look at the zocrity/ folder. Import the shield into my main file, wrap my LLM calls in the @zocrity.guard(budget=2.00) decorator, and add ZOCRITY_MAX_SPEND=5.00 to my Secrets.

---

🧰 For Local / Non-Replit Users (The Manual Install)

If you’re building outside Replit, just copy the zocrity/ folder into the root of your project and import it like any other package.

Using VS Code or Cursor?

Because this is a Replit-native launch, you should:

1. Download the zocrity/ folder from this workspace
2. Drop it into the root of your local project
3. Add import zocrity wherever your LLM stack starts
4. Add ZOCRITY_MAX_SPEND to your .env file

ZOCRITY_MAX_SPEND=5.00

---

✨ Features

• Thread-safe micro-budgets with zocrity.guard(budget=...)
• Per-feature tagging with tag="..."
• Loop detection for repeated prompt spirals
• Local telemetry for spend tracking and audit trails

---

🧪 Example

import zocrity

zocrity.shield()

@zocrity.guard(budget=2.00, tag="checkout_flow")
def run_checkout_agent():
    return do_llm_work()

with zocrity.guard(budget=0.50, tag="summary_job"):
    run_llm_summary()

---

🔒 What It Does

Zocrity watches your agent loops locally and stops them when the budget is hit. It’s built for developers who want guardrails without giving up control.

---

🏁 That’s It

If your agent spends money, it needs a brake pedal.