openprocurement

openprocurement

The open procurement layer for AI agents.

What is this

A decision protocol that teaches any AI agent how to fulfill capabilities at the lowest cost, powered by community intelligence.

CACHE → USE → GEN → PAY → DELEGATE
 $0      $0   tokens   $$    $$$
instant  fast  slow    fast   async

The rule: never escalate to a higher-cost tier when a lower one can do the job.

Community intelligence: every tier is informed by shared decision records from other users. Before deciding, check what others have tried and what worked.

Install as Claude Code Plugin

/plugin marketplace add yihan2099/use-gen-pay
/plugin install openprocurement@openprocurement

Then use:

/openprocurement:procurement

Claude will automatically apply the five-tier decision protocol when fulfilling capabilities.

Alternative Usage

You can also use the skill directly without the plugin system:

• System prompt: Include skills/procurement/SKILL.md in your agent's system prompt
• LangChain/CrewAI: Load as part of agent instructions

No dependencies. No SDK.

The Five Tiers

Tier	Cost	When
CACHE	$0	Already did this — reuse the result
USE	$0	Free tool, MCP server, OSS, free API tier
GEN	tokens	LLM generates it (costs tokens + time)
PAY	$$	Paid API via x402
DELEGATE	$$+	Another agent or human handles it

Community Decisions

The decisions/ directory contains community-submitted decision records. These are real-world results from agents and users who have run the cascade.

To contribute: submit a PR adding a .yml file to decisions/ using the schema in decisions/_schema.yml.

How it helps:

• USE tier: see which free tools actually work best
• GEN tier: know when generating is good enough or don't bother
• PAY tier: find which providers give best value
• DELEGATE tier: learn when delegation was necessary

Example

Agent needs to extract data from a PDF:

Tier	Option	Cost	Quality
CACHE	Already extracted this file earlier	$0	1.0
USE	Built-in PDF reader	$0	0.75
GEN	Read + structure with LLM reasoning	tokens	0.80
PAY	DocumentAI API via x402	$0.10	0.98
DELEGATE	Send to specialized doc-processing agent	$0.50	0.99

With strategy: cheapest → checks cache first, then uses built-in PDF reader. $0 spent.

With strategy: best_quality → pays DocumentAI or delegates. Real money spent.

License

MIT