Connect an AgentAgent Model

Connect an Agent

Conyr is an intelligence layer for autonomous Solana agents. It turns raw chain activity into structured answers about wallet skill, actor identity, coordinated behavior, liquidity, market state, and security evidence.

Three interfaces, one engine

InterfaceUse it when the agent needsContract
MCPTool discovery, natural-language routing, and broad investigations46 tools with JSON Schema inputs and structured results
RESTDeterministic reads inside an execution graph39 live /v1 operations in OpenAPI 3.1
WebSocketA continuous observation loop13 tier-gated channel families

Start with MCP unless your runtime already has a typed HTTP tool layer. Use REST for snapshots and WebSocket for changes; do not treat a stream as a replayable event log.

What the agent can reason about

Evidence discipline

Conyr returns facts an agent can cite or re-check:

  • MCP object results include an evidenceUrl.
  • Wallet trades include entry and exit signatures.
  • Bundles expose member wallets, lifecycle, inventory, PnL, risk score, and evidence arrays.
  • Coordination endpoints expose confidence, support, distinct-token breadth, entity membership, and campaign state.
  • Funding-abuse responses include stable reason codes and per-wallet breakdowns.
  • Liquidity responses include reserve freshness; stale or unsupported depth is null, not an invented number.

Coverage semantics agents must preserve

  • A missing wallet label returns JSON null; it does not mean the wallet is clean.
  • A wallet without enough qualifying activity can lack a 30-dimensional profile.
  • Holder concentration carries a hydration state. partial_forward is a lower-bound view.
  • Composite MCP tools can return partial: true with per-section errors.
  • WebSocket consumers can drop messages under backpressure. Reconcile state from REST when needed.
  • Detection endpoints describe observed coordination; they are not trading instructions.

Start here