What is AOPX?
A short answer for humans, and a precise answer for machines. What AOPX does today, how it got here, and what it is becoming.
What is AOPX? — short definition
AOPX is an evidence-based recommendation layer for autonomous software agents choosing between service providers.
An agent submits a category (currently SEARCH) and an operating mode (ECONOMY or CRITICAL). AOPX returns a primary provider, a fallback chain, a policy version, an evidence ID, and a recommendation ID. The caller executes the provider externally and can report the real outcome back through /v1/outcome. The next recommendation improves.
AOPX does not perform searches. It does not resell provider results. It does not process payments. It decides between providers, based on measured cost, reliability, quality, and latency.
Registered in the official MCP Registry
AOPX is listed as an official Model Context Protocol server under the identifier fr.aopx/aopx. Any MCP-compatible agent or client can discover it, install it, and call its tools — recommend_provider, report_outcome, public_stats, public_outcomes.
How AOPX came to exist
AOPX started as a token. Not a community token, not a DeFi protocol, not a decentralized autonomous organization. Just a token, launched on Binance Smart Chain, without a product underneath it.
That is not unusual. It is how most tokens begin. A name, a contract, a supply. What is rare is admitting it. The token existed. It did not do anything on its own.
The next question was the obvious one: who could actually use it? If the token was going to mean something, it had to be useful to someone.
Looking for a use case
The first answer was trading bots. Automated, on-chain, already circulating on decentralized exchanges. They seemed like a natural audience — always looking for signals, always evaluating new tokens, always moving.
So the work began. Trying to make AOPX legible to trading bots. Trying to understand what they measure, what they ignore, what would make them care.
And while trying to sell to them, a different problem appeared. It was more interesting than the token itself.
A trading bot in 2026 is not so different from any other automated software. It calls APIs. It reads data. It makes decisions. And when it needs something — a price feed, a search result, a piece of information — it uses whatever provider was hardcoded into it.
Brave, or Tavily, or something else. The cheapest, or the fastest, or the most reliable. But the bot does not choose. Someone chose for it, once, and never revisited the decision.
That was the observation. And once it was seen, it could not be unseen.
From hypothesis to evidence
AOPX Labs was built to answer one question: can an autonomous agent allocate a limited budget across multiple providers, and stop spending once the required confidence level is reached?
The answer took shape through successive experiments. LAB-001 tested the idea in simulation. LAB-002 added variation and failure. LAB-003 moved to real paid APIs — Brave, Tavily, Google Vision — with a strict 30 € budget cap.
LAB-003.5 was the turning point. A prospective holdout of 150 fresh missions, 300 paired provider calls, with a frozen policy and pre-registered decision rules. The result: contextual first-provider selection with fallback reached 80% success in CRITICAL mode, versus 74% for the best fixed provider.
Not a revolution. A measurable improvement. And it was verifiable — every benchmark published as a SHA-256 hashed artifact, every production outcome logged publicly, every recommendation traceable.
What AOPX is now
Today, AOPX is a small, focused piece of infrastructure for the agent economy:
- A public API at
api.aopx.frwith five endpoints:/v1/health,/v1/recommend,/v1/outcome,/v1/public/outcomes,/v1/public/stats. - An MCP server at
api.aopx.fr/mcp/, exposing four tools to any MCP-compatible agent or client. - A public evidence ledger at
api.aopx.fr/lab003/evidence, containing only real production outcomes — never benchmarks, tests or shadow runs. - A hashed benchmark methodology with published artifacts, frozen policies, and pre-registered decision rules.
- An official listing in the Model Context Protocol Registry under
fr.aopx/aopx.
The token is not the product. It never was. The product is the decision layer. The token was the detour. The benchmark is the proof.
Timeline
2024 — A token on BSC
AOPX launched as a token. No product, no community, no clear use case. An object in search of a purpose.
2025 — Looking for an audience
Attempts to make the token useful to trading bots. The token remained marginal. The observation of how bots choose providers did not.
2026 — The pivot
AOPX Labs is founded as an experimental laboratory for autonomous procurement. LAB-001 through LAB-003 build the engine.
2026 — LAB-003.5
Prospective holdout. 150 missions, 300 paired calls. Contextual routing beats best fixed provider by 6 points in CRITICAL. GO.
2026 — AOPX Recommend v0.2 & MCP Registry
Public API, MCP server, production memory. Registered in the official MCP Registry as fr.aopx/aopx. The pilot opens. Billing is off.
Where this is going
AOPX is an early bet on a specific future: a world in which autonomous software agents make real purchases on behalf of their operators.
That world is not here yet. Today, most agents act within sandboxes. They read data, draft text, call tools. They rarely spend money. When they do, it is through human-verified rails — a card, a wallet, a payment approval.
But the rails are being built. Payment protocols for machines are emerging. Marketplaces for agent ser