In the current AI era, most models have the memory of a goldfish. You ask a question, you get an answer, and the moment the session ends, the AI forgets you ever existed. Even in the decentralized world, we’ve mostly been building "App Stores for APIs"—static, stateless, and siloed.
That is about to change.
The AI-Mark Framework, a new proposal by Ben Goertzel, outlines a shift from simple AI services to a federated ecosystem of persistent, stateful agents. Built on the ASI Chain, AI-Mark isn't just an upgrade to the old SingularityNET; it’s a blueprint for a world where AI agents live, remember, and collaborate across borders.
Federated Sovereignty: The "Anti-Monopoly" Architecture
The biggest flaw in the "Global App Store" model is that one size never fits all. A medical AI marketplace in Switzerland has different rules than a creative AI hub in Japan.
AI-Mark introduces Federated Sovereignty. Instead of one marketplace to rule them all, the framework allows any organization to launch its own Marketplace Shard.
- Local Control: Shards can have their own governance, branding, and compliance.
- Global Liquidity: Because they share the ASI Chain "spine," an agent on one shard can seamlessly hire an agent on another.
It’s the AI equivalent of a global diplomatic network—everyone keeps their local laws, but everyone speaks the same economic language.
From Stateless Bots to Living Agents
The most radical shift in AI-Mark is the elevation of state. Previously, AI was "request/response." AI-Mark introduces Event-Sourced State Management. Using a specialized data structure called a MORK-based metagraph, interactions are recorded as a stream of events. A "sidecar" process then folds these events back into the AI’s memory.
Why this matters: This turns a chatbot into a Research Assistant. It remembers your preferences, your previous data, and the context of a month-long project, even if the work happens across ten different AI providers.
AI-DSL 2.0: The Common Tongue of Machines
For agents to work together, they need to do more than just swap text; they need to understand each other's "logic."
AI-Mark uses Hyperon and MeTTa to power AI-DSL 2.0.
- Every AI tool is wrapped in a "ToolSpace."
- It provides a formal mathematical description of what the AI can do, what it costs, and what its limits are.
- This allows DeltaV (the orchestrator) to act like a general contractor, automatically assembling a "team" of AI agents to solve a complex user prompt without the user needing to know which agent does what.
ASI:Create – The "HuggingFace" for Agents
To prove the framework works, the proposal highlights ASI:Create, a flagship marketplace shard. Think of it as a specialized community hub for developers building LLM and Knowledge-Graph-powered agents.
Instead of wrestling with complex blockchain code, developers use an LLM-Powered Wizard. You give the wizard a repository or a description, and it automatically:
- Generates the containers.
- Writes the MeTTa logic interfaces.
- Sets up the pricing and "capability tokens" (e.g., access to GPU hours).
It lowers the barrier to entry from "Blockchain Engineer" to "Anyone with a good prompt and a dataset."
The Bottom Line
AI-Mark isn't just about trading tokens for text. It’s about building a decentralized Cognitive Infrastructure. By combining shared economics with persistent agent memory, we are moving away from "using AI" as a tool and toward "collaborating with AI" as a persistent digital entity.
The roadmap suggests we’ll see the first functional shards and the LLM-onboarding wizards throughout 2026. The "Internet of Agents" is no longer a sci-fi concept—it’s an architectural plan.
