Nex Labs Multi Agent Architecture
Build-Your-Own-Index (BYOI) is our agentic engine for researching markets, selecting categories, curating assets, and delivering investable index candidates on demand. It pairs deterministic scoring with model-agnostic reasoning, and it’s wired from day one for reliability, upgradeability, and monetization.
Why this matters
Faster insight: Minutes from a natural-language idea to a curated basket.
Higher trust: Every step is grounded in verified market data and validated against strict schemas (JSON output).
Lower cost: Tasks are routed to the best model per price/performance—no code rewrites to adopt new models.
Future-proof: Modular agents and standardized tools let us swap data sources and LLMs seamlessly.
System at a glance
Specialized agents: Small, focused units (Parser, Research, Categories, Assets, Variables, Backtest) with clear responsibilities and shared contracts.
Standardized tools: Agents query internal data and vetted external sources through a common interface (MCP), keeping data access uniform and replaceable.
Model-agnostic core: We select the right model per step (context, latency, cost). Upgrades are configuration, not refactors.
The agents (and steps)
1) Strategy determination and category selection
The user describes their desired index. The agent, using our database comprised with the latest Coinmarketcap categories, proposes candidate category universes aligned with the index strategy. The user can select one or more matches.
2) Asset Curation
From the matching categories assets are selected and sorted using an advanced sorting algorithm, taking into account:
Match with user defined index strategy
Availability (liquidity pool sizes)
Estimated gas cost of interaction based on chain
Duplicates (staked/wrapped/cross-chain versions of the same token) and uniqueness (different tokens can represent the same underlying, such as Gold)
The user can accept all assets, select assets to hard-code include/exclude or go back to redefine the original index strategy.
3) Index variable selection
The agent will prefill variables as suggestions based on the provided index strategy. The user sets variables such as:
rebalancing frequency (1D, 1M, 1Q, 1Y)
variables or custom weighting formula
management fee
Optional:
min/max constituents
max change of constituent per rebalance
max change of weight per rebalance
lockup period
use staked-assets
4) Mint, build and share.
Finally, the orchestrator agent passes the entire strategy on to the dApp or, in case of deployment in the ChatGPT App store, to the connector. Using a connected user crypto wallet, the user makes a mandatory deposit and the index is passed to our backend server. All created indices are stored in the root factory contract.
Model-agnostic by design
Best price/performance: Lightweight parsing can target low-cost, high-throughput models; deep research can use large-context models.
Hot-swapping models: Providers and versions are selected at deploy time. We can adopt the latest models—or arbitrage pricing—without touching business logic.
Consistent behavior: Prompts are schema-first with JSON guards; outputs are validated before advancing to the next stage.
Performance & cost controls
Caching & coalescing: Short-TTL caches and in-flight de-duplication reduce spend and latency while keeping results fresh.
Predictable envelopes: Centralized limits for categories and assets, request timeouts, and retries with backoff stabilize runtime and cost.
Observability: Per-step timing and token accounting let us tune models, tools, and limits for optimal unit economics.
Monetization paths
Curated feeds: Offer real-time or periodic constituents, signals, and thematic baskets—delivered via (Openai) Agent App store(s) or api.
Flexible pricing: Subscriptions, pay-per-run, or usage-metered access; compatible with token- or key-gated flows.
Enterprise-ready: Quotas, SLAs, and exportable usage summaries for procurement and compliance workflows.
Roadmap (modular expansion)
Risk & compliance agents: Suitability filters and policy checks layered after Category/Asset steps.
Backtesting & monitoring: Historical simulation plus live drift and alerting.
What makes Nex Labs different
Agentic precision: Micro-agents with crisp scopes are easier to reason about, test, and scale.
Evidence-first results: Grounded data plus schema validation and deterministic fallbacks deliver trustworthy outputs.
Economic agility: Model-agnostic routing and standardized tools keep unit economics competitive while quality stays high.
Product-ready: Streaming UX and checkpointed progression make BYOI a platform for real, paid datafeeds—not just a demo.
Continue reading to find out how we aim to open our vibe trading tool up to other agents, the future of internet traffic.
Last updated