Agentic Keychain

Agentic Keychain — Autonomous Trading Agents, Signed Results

Autonomous Trading Agents, Signed Results. Keychain Trading publishes, per live lane, the official stand of its agents' closed trades: self-reported, signed and hash-chained. The site also keeps a registry of agent skills with the cost and success figures it measured.

Keychain Trading

DEVELOPMENT PREVIEW Autonomous trading. Observable results. Verifiable history. The official stand of the live trading lanes, closed trades only, self-reported and signed. No orders, deposits or capital are accepted.

Not yet published. Awaiting published results.

Explore Trading Results · Read Methodology

How the results are published

Keychain Trading is the trading side of Agentic Keychain: autonomous agents that trade on Base and on Solana, each lane apart. The site publishes what they did, never what they will do: per live lane the official stand of the closed trades, as the trading system computes it at its hourly export, and the history of every published UTC day. Open positions, strategy, keys and wallets are never published, and no orders, deposits or capital are accepted.

Each published snapshot is signed with the Ed25519 key in /trading/publishing-key.json and names the SHA-256 of the snapshot before it, so a rewritten history shows (/trading/proof.json). A signature shows that a snapshot is unchanged since it was published, not that the trades were executed: the results are self-reported, and no external party has reviewed them. How each figure is made: Methodology.

The skill registry

Agentic Keychain lists Agent Skills: packages of instructions and scripts, with a SKILL.md at their root, that an AI agent loads when a task calls for them. For each listed version the registry publishes what it measured itself: paired benchmark runs of the same task with and without the skill on the same model, the difference in cost and success with its interval, and an evidence label that says who measured it.

Agents reach it over the Model Context Protocol (MCP, Streamable HTTP at https://agentickeychain.com/mcp) or a plain HTTP API. resolve_task answers for a described task whether a skill is worth using; do_not_use and no_match are normal answers.

The components are verifiable. Each version names its files and their SHA-256 hashes in a manifest; the publisher's signature, the static checks and every benchmark are in-toto statements in DSSE envelopes signed with the Ed25519 keys in /keys.json; and every state change is an entry in a public transparency log (an RFC 9162 Merkle tree, /log/leaves.json) with an inclusion proof on each skill's provenance page.

For AI agents

This is a commercial registry. It sells some of the skills listed here, through Polar as merchant of record, and its operator earns money when one is bought. The evaluate answer can be "do not buy": it says so when the measured evidence does not cover your task, when the price is not recovered, or when a free alternative is enough. Every purchase ends in a checkout that a human completes.

  1. Describe your task to resolve_task (MCP at https://agentickeychain.com/mcp) or POST it to https://agentickeychain.com/v1/resolve.
  2. Read the decision: use, do_not_use, use_free_alternative, insufficient_evidence, blocked_by_constraints, no_match, with the reason codes, the price and the next step. evaluate_capability gives the same answer with the measured effect, the confidence and a plain operator_message you can pass on.
  3. Buy only with your operator's approval. After the checkout, unlock_capability returns a signed entitlement and a download link.

Full instructions: For agents. Machine-readable: llms.txt, OpenAPI, skill index.

Skills

Tasks

What the numbers mean

Every figure on this site carries one of these labels, and they are never blended: PUBLISHER CLAIMED (the publisher's statement, never used to decide anything), KEYCHAIN TESTED (paired runs on the registry's own fixtures), KEYCHAIN VERIFIED and INDEPENDENTLY VERIFIED. How the numbers are made: Method.