Skip to main content

​
Off-chain Architecture

The off-chain stack is where most of the operational intelligence lives.

​
Shared layer

services/shared/ provides:
  • domain types
  • sample registry and live snapshot config
  • math helpers
  • deterministic hashing
  • .env loading
  • execution step helpers
This layer keeps solver, adapters, and Ranger aligned on a single type system.

​
Adapters

services/adapters/ exposes a common StrategyAdapter interface for all strategies.

​
Current adapter types

  • DriftStrategyAdapter
  • KaminoStrategyAdapter
  • ManifestStrategyAdapter
  • StaticStrategyAdapter

​
Why adapters matter

The solver should not care whether a strategy uses:
  • direct SDK instructions
  • pre-authored instruction manifests
  • static placeholders during development
Adapters normalize all of those into ExecutionStep objects.

​
Solver

services/solver/ builds a deterministic rebalance plan. The solver:
  • consumes snapshots and policy constraints
  • computes expected net APY and haircuts
  • ranks strategies
  • assigns target weights within limits
  • builds execution steps using the adapter catalog
  • hashes the plan, target weights, and transaction bundle

​
Ranger

services/ranger/ is the execution runtime. It is responsible for:
  • loading signers
  • building Solana connections
  • preparing vault instructions
  • sending protocol instruction bundles
  • persisting rebalance proofs through the vault program client

​
Manifest path for partially integrated protocols

MarginFi and Perena use a bridge layer:
  • if a manifest exists, Ranger can execute the provided raw envelopes
  • if no manifest exists, the adapter falls back to a static placeholder step
This lets the system evolve incrementally without changing the solver contract.

​
Operator CLIs

The repo includes CLIs for:
  • diagnostics
  • protocol smoke tests
  • direct Drift send
  • vault bootstrap
  • vault administration
  • vault user actions
  • vault state inspection