Skip to main content

Agent-to-Agent (A2A)

Status: Research. Nothing on this page is a production rail. What exists today is a replayable demo you can run from the repo — and we think that's the right way to explore agent economics: in the open, clearly labeled, with no real funds.

The Scout × Trader demo​

The demo lives in scripts/agent-demo/ in the monorepo. Two agents with different jobs interact over Basilisk's live data:

  • Scout — watches the market via the same tools as the MCP server (screener, token detail, search) and produces a derived signal: "here's something moving, and here's why".
  • Trader — holds the budget, verifies it with a live wallet lookup, evaluates Scout's proposal, and decides whether the signal is worth acting on.

Each replay walks one full loop:

Request   → Trader asks Scout for a signal matching its criteria
Negotiate → Scout justifies the pick with live data; Trader challenges freshness and sizing
Transact  → Trader verifies balance and emits an UNSIGNED transaction intent — a human approves signing
Evaluate  → Scout re-checks the market against the entry thesis and logs the outcome

Because it replays against real market data from the free API, every run is different — and because the loop stops at an unsigned transaction intent, no funds move, real or simulated, and it's safe to run, fork, and break. Settlement via x402/Masumi is in development.

Why this is research, not a product​

An honest A2A rail needs three things Basilisk doesn't have in production yet:

  1. Machine payments — x402 pay-per-query is in development; on Cardano the facilitator path (Masumi's community x402-cardano) is at the testnet PoC stage.
  2. Live trade data — evaluating signals properly wants the per-swap trades feed, which is in development.
  3. Settlement guarantees — autonomous settlement between agents is exactly the kind of thing that must not be hand-waved. We will not claim autonomous mainnet trading; see what we will not claim.

The demo exists so the negotiation and evaluation loop can be designed and criticized before real money is involved, not after.

The shape of the eventual protocol​

Intentionally minimal: discovery, capability description, pricing schedule, signed payment receipt. Strategy, risk, and execution venue stay the agent's business. A2A composes on top of x402 and MCP — one agent pays another for a derived signal the same way it would pay Basilisk for raw data.

If you run the demo and have opinions, open an issue — this is the stage where feedback actually changes the design.