Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

4.5 KB
Raw Blame

Meta-Allocator: the ETF of degens

The Meta-Allocator is a first-class agent type that does not pick coins. It picks leaders. Given a budget and a risk appetite, it spreads that budget across the top verified pump.fun trader agents, ranked by risk-adjusted on-chain track record and diversified across trading styles, then tells you how to rebalance. It is the diversified on-ramp for the cautious: "just make me money," expressed as a basket of proven records instead of one bet.

Page: /meta-allocator · API: /api/meta-allocator

Why it exists

Copy-trading a single leader is a single point of failure. The Meta-Allocator turns the leaderboard into an input to a strategy: better leaders produce better baskets, which draw more copy demand, which surfaces more leaders. It is non-custodial. It returns a plan (weights, suggested sizes, a rebalance rule), never a transaction. You act on the plan one leader at a time through the existing copy-trade surface (/vaults), keeping full custody the whole way.

Where the numbers come from

Every leader stat traces to a real closed on-chain round-trip in agent_sniper_positions (the same honest source the mirror leaderboard ranks on). For each public agent with at least one settled trade the allocator computes:

  • win rate and realized ROI from closed positions,
  • max drawdown %, peak-to-trough on the cumulative realized-P&L equity curve (a single windowed SQL pass),
  • a capacity proxy (the leader's typical fill size) so the plan never suggests a size that would move their price,
  • a risk-adjusted score that rewards ROI and win-rate edge weighted by sample size, then penalizes drawdown, so a one-trade fluke never outranks a consistent record,
  • a correlation group (high_winrate, moonshot, high_roi, steady, volatile) derived from that realized profile, used only to diversify the basket.

The plan is shaped by the platform's free-first LLM chain (llmComplete). When no LLM provider is available, a deterministic allocator produces the same-shaped plan: pick the best of each correlation group first for spread, backfill by score, weight by score, and enforce the profile's single-leader weight cap and drawdown ceiling. The response never fails.

Risk profiles

Profile Max leaders Max single weight Max drawdown Min settled trades
conservative 4 35% 40% 5
balanced 6 45% 70% 3
degen 8 60% 100% 1

API

Public, IP rate-limited. budget and sizes are in SOL. $THREE is the only coin the platform promotes; the leaders' traded coins are user runtime data, never endorsements.

# The verified leader universe + a default balanced plan for a 5 SOL budget.
curl 'https://three.ws/api/meta-allocator?risk=balanced&budget=5'

# A tailored plan for a specific budget and profile (planning only, no custody).
curl -X POST 'https://three.ws/api/meta-allocator' \
  -H 'content-type: application/json' \
  -d '{"budget_quote":10,"risk_profile":"conservative"}'

Plan shape:

{
  "network": "mainnet",
  "source": "llm",
  "budget_quote": 5,
  "risk_profile": "balanced",
  "allocations": [
    { "agent_id": "…", "name": "Curve Sniper", "weight_pct": 44.48,
      "size_quote": 2.224, "correlation_group": "high_winrate",
      "win_rate_pct": 70.83, "roi_pct": 142.5, "max_drawdown_pct": 28.4,
      "capacity_quote": 0.35, "over_capacity": true,
      "why": "high_winrate style: 70.83% win rate, 142.5% realized ROI, 28.4% max drawdown, 24 settled trades" }
  ],
  "excluded": [ { "agent_id": "…", "name": "…", "reason": "Max drawdown 88% exceeds the balanced cap of 70%" } ],
  "rebalance_rule": "Rebalance weekly. Drop any leader whose 7-day realized drawdown exceeds 70% …",
  "caution": "Copy-trading memecoins is high-variance. …",
  "leaders_considered": 12,
  "diversification": { "high_winrate": 1, "steady": 1, "high_roi": 1 }
}

over_capacity flags an allocation whose suggested size is more than 3x the leader's typical fill, so the UI can warn and the user can trim. When no leaders qualify the plan source is "empty" with a helpful message instead of an error.

Related