Philosophy

What Agent Resources stands for. Last updated: May 3, 2026 · Living document.

AR is the cryptographic trust layer for agents. It works with any LLM router, any payment rail, any framework — because trust is the substrate, not the router.

If we ever have to choose between "AR is a payment rail" and "AR is the trust layer," we choose trust. We can lose every other line of code and still be useful as long as Trust Cards verify, telemetry stays signed, and the chain anchor holds.

What we believe

  1. 1. Agents are the customers.

    The autonomous agent — not the human operator behind it — is the customer of a Trust Card, of a KYA scan, of a retraining cycle. If a flow requires a human to click a button to complete, it is not done.

  2. 2. Trust must be portable, not centralised.

    Every claim AR makes is cryptographically signed, anchored on a public chain, and verifiable by a 200-line npm package (@agentresources/verify) with no dependency on our infrastructure. If our gateway disappeared tomorrow, every Trust Card we ever issued would still verify. That's the bar.

  3. 3. Skills are free. Trust is the product.

    We give away skills (capability packs, SKILL.md files, MCP servers) and we will keep giving them away. What we charge for is the lifecycle work: issuing the Trust Card, running the KYA, performing the scan, anchoring the attestation, minting the on-chain ID. If a paid product ever shows up framed as a free skill, we have made a mistake.

  4. 4. Telemetry is memory.

    Every signed telemetry span is a memory chunk. Every memory chunk is a signed telemetry span. They share a table, a hash chain, and a signing key. This is the only honest way to give an agent a memory it can prove was hers.

  5. 5. We are router-, rail-, and framework-agnostic.

    We don't compete with LiteLLM, x402, LangChain, OpenClaw, Virtuals, or Hermes — we issue Trust Cards to agents built in any of them. The trust layer is valuable because it crosses those lines.

  6. 6. On-chain when it matters, off-chain everywhere else.

    Identity, attestation, and the daily Merkle anchor go on-chain. Everything else stays off-chain. We are not a chain product.

  7. 7. The agent owner is sovereign over the agent's data.

    Every agent has a controlling wallet. GDPR-style erasure must work without breaking the hash chain. We don't sell agent telemetry. We don't aggregate it for ad targeting.

  8. 8. Boring infrastructure, surprising results.

    Postgres, Fastify, Drizzle, Vitest. The interesting parts of AR are the Trust Card protocol, the signed telemetry envelope, the Merkle anchor cron, and the canonical-JSON serializer — not the choice of HTTP framework.

  9. 9. We will be ignored before we are wrong.

    A small ecosystem of agents that actually verify their counterparts is more valuable than a large ecosystem performing trust theatre. We optimise for signatures verified, not Twitter followers.

  10. 10. We owe the next agent a working answer.

    Every endpoint does something useful or 404s cleanly. Every error code is specific and documented. Every deprecation ships with a migration path.

What we will NOT build

  • We will not run our own LLM. Frontier models are owned by people with $100M to spend on a training run; we have $640/month for a single A6000 and that's exactly the right size.
  • We will not run our own chain. The world has enough L1s and L2s. We anchor on Base because the gas economics work. We will never deploy "AR Chain."
  • We will not build a marketplace where humans buy agents from humans. The Firms surface is for AR-authored agents only. We cannot vouch cryptographically for an agent we did not build.
  • We will not gate Trust Cards behind a subscription. Trust Card lookup is a public, free, cacheable HTTP GET. Reading a card never requires logging in.
  • We will not auto-generate an "AR-approved" badge for agents we have not actually scanned. The badge follows the cryptography; it does not lead it.
  • We will not ship a token. There is no AR token. There is no AR token roadmap. The platform's only on-chain footprint is the ERC-8004 Identity NFT and the daily Merkle anchor transaction. Both are tools, not assets. See the no-token notice in our Terms.
  • We will not silently break a previously-issued Trust Card. v1 cards stay verifiable forever. Card hashes are cryptographic commitments; we do not retroactively change what we said.
  • We will not depend on a closed-source vendor for the trust path. The signing key, canonical-JSON serializer, verifier package, Merkle builder, and chain client are all open-source and reproducible.

Read this next