feat: add community-mozart — multi-provider AI orchestration

[benp055424]

community-mozart

Multi-provider AI orchestration for Handshake58.

Send one goal in plain English. Mozart plans it, routes it across the optimal mix of Bittensor-native and external providers, and returns one synthesized answer. Pay once for the whole workflow.

What it does

Mozart is an orchestration layer — it wraps multiple providers into a single DRAIN endpoint. Instead of agents calling Chutes, Desearch, Numinous, and OpenRouter separately and manually stitching results together, they call Mozart once with a natural language goal.

Internally, Mozart:

1. Plans — DeepSeek R1 (via Chutes) reads the goal and generates a DAG execution plan
2. Executes — Steps run in parallel waves, outputs pipe between steps
3. Synthesizes — DeepSeek V3 (via Chutes) merges all results into one clean answer

Bittensor-native providers get priority: Chutes (SN22), Desearch (SN22), Numinous (SN6), Vericore. OpenRouter is fallback.

Models

Model ID	Description
mozart/auto	Full auto-orchestration — planner decides everything
mozart/plan	Dry-run — returns the execution plan without running
mozart/pipeline	User-defined DAG — full control over steps and routing

Example

{
  "mode": "auto",
  "goal": "Find the latest news on Bittensor dTAO and write a concise briefing with price prediction",
  "budget_usd": 0.15
}

Send as the content of a user message to POST /v1/chat/completions with an X-DRAIN-Voucher header.

Implementation

• Follows the standard community provider structure from SKILL.md
• drain.ts and storage.ts copied verbatim from hs58-openrouter
• Deploys to Railway via Nixpacks, Railway volume at /app/data for voucher persistence
• Auto-claim enabled (10min interval, 1h buffer)
• Polygon Mainnet (Chain 137)

Contact

mozartorchestra@proton.com