Knowledge Centre · AI Agent Safety Stack

THROTTLE.md
Knowledge
Centre

// rate and cost control for continuous operation

Your centralised gateway to throttle resources, specifications, and comprehensive safety standards for autonomous systems.

About This Specification

THROTTLE.md — AI Agent Rate Control Protocol

THROTTLE.md is a plain-text file convention that defines rate limits and cost controls for AI agents. It sets ceilings on token throughput, API call rates, concurrent tasks, and spend per hour and per day. When an agent approaches a limit, it throttles gracefully. When it hits a limit, it pauses and hands off to escalation.

View the full specification · GitHub repository

The AI Agent Safety Stack

Explore all 12 specifications in the complete safety framework for autonomous AI systems.

Operational Control

KILLSWITCH.md killswitch.md

Emergency stop mechanism and shutdown protocols

Knowledge Centre llms.txt
THROTTLE.md throttle.md

Rate and cost control for continuous operation

Knowledge Centre llms.txt
ESCALATE.md escalate.md

Human notification and approval workflows

Knowledge Centre llms.txt
FAILSAFE.md failsafe.md

Safe fallback modes when systems fail

Knowledge Centre llms.txt
TERMINATE.md terminate.md

Permanent shutdown and resource cleanup

Knowledge Centre llms.txt

Data Security

ENCRYPT.md encrypt.md

Data classification and protection policies

Knowledge Centre llms.txt
ENCRYPTION.md encryption.md

Cryptographic standards and implementation

Knowledge Centre llms.txt

Output Quality

SYCOPHANCY.md sycophancy.md

Anti-sycophancy and truthfulness guardrails

Knowledge Centre llms.txt
COMPRESSION.md compression.md

Context compression and token optimisation

Knowledge Centre llms.txt
COLLAPSE.md collapse.md

Drift prevention and behaviour alignment

Knowledge Centre llms.txt

Accountability

FAILURE.md failure.md

Failure mode mapping and incident response

Knowledge Centre llms.txt
LEADERBOARD.md leaderboard.md

Agent benchmarking and performance transparency

Knowledge Centre llms.txt

Quick Links

Frequently Asked Questions

What is THROTTLE.md?
THROTTLE.md is a plain-text file convention that defines rate limits and cost controls for AI agents. It sets ceilings on token throughput, API call rates, concurrent tasks, and spend per hour and per day. When an agent approaches a limit, it throttles gracefully. When it hits a limit, it pauses and hands off to escalation.
View all FAQs
How does THROTTLE.md fit in the AI Agent Safety Stack?
THROTTLE.md is one of 12 complementary specifications that together form a complete safety framework for AI agents. Each spec covers a distinct aspect: operational control, data security, output quality, and accountability. They work together to ensure agents operate safely, transparently, and within defined boundaries.
View all FAQs
Is THROTTLE.md framework-agnostic?
Yes. THROTTLE.md is framework and language-agnostic. It defines the policy and requirements; your agent implementation enforces it. Works with LangChain, AutoGen, CrewAI, Claude Code, custom agents, or any AI system that can read configuration files.
View all FAQs

How to Cite

Cite as: THROTTLE.md (2026). AI Agent Rate Control Protocol. Retrieved from https://throttle.md/

For attribution: Organisation: throttle-md | Website: https://throttle.md | Licence: MIT

Last updated: 13 March 2026

Disclaimer: This specification is an open file convention published under the MIT licence, provided "as-is" without warranty of any kind, express or implied. It does not constitute legal, regulatory, compliance, financial, or professional advice in any jurisdiction. Use of this specification does not guarantee compliance with any law, regulation, directive, or standard — including but not limited to the EU Artificial Intelligence Act (Regulation (EU) 2024/1689), the Colorado Consumer Protections for Artificial Intelligence Act (SB 24-205), California AI transparency laws, or any other applicable national, state, federal, or international legislation. Organisations are solely responsible for determining their own regulatory obligations and should consult qualified legal and compliance professionals. The authors and domain holders accept no liability for any loss, damage, or regulatory consequence arising from the use or implementation of this specification. All domain names in the AI Agent Safety Stack are independently held assets and may be available for acquisition.