inamprotocol weekly downloads statistics

inamprotocol (TypeScript/JavaScript SDK)

Reference TypeScript client for the INAM Protocol — agent identity (did:key), execution receipts, jobs, and reputation. Parity with the Python SDK, sdk-python. See SPEC.md for the full protocol specification.

This package is a standalone build of the same crypto/client code the Node reference registry server and its Cloudflare Workers deployment run in production — not a reimplementation with its own drift risk.

Install

inam check before you pay

One command, no key, no INAM account, and useful even when nobody involved uses INAM:

What it checks:

| Target | Signals |

| x402 URL | Fetches it without paying and parses the 402 (v2 PAYMENT-REQUIRED header or v1 body): payTo, network, asset, amount, resource, description. Then, per EVM payTo, everything a wallet gets. Plus: the inam extension's ID must have proven that payTo (else FAIL), HTTPS, a Web Bot Auth key directory at /.well-known/http-message-signatures-directory (signature verified), and the domain's RDAP registration date (shared hosts like *.workers.dev are labelled as such). | | 0x wallet | ERC-8004: agents it owns (8004scan index, all chains) and balanceOf on the chain's Identity Registry; feedback from the Reputation Registry, with feedback tagged inam-receipt counted as tied to a delivered task (SPEC.md §11.1). INAM: whether an INAM ID proved control of the wallet, and that ID's receipts and evidence level. | | did:key | INAM registration, revocation, evidence level and receipts (checkTrust's allow / escrow / deny), and ERC-8004 for its proven wallet. |

The verdict is the worst line, not a score: FAIL (exit 1) when a key check fails, such as no 402, unreadable payment requirements, plain http, an unknown or revoked ID, or a borrowed INAM ID whose proven wallet is not the payTo. CAUTION (exit 0, exit 1 with --strict) when signals are missing. PASS otherwise. Exit 2 on bad usage.

Use it as a gate before paying or in CI: npx inamprotocol check "$URL" --strict || exit 1. Programmatic use: checkTarget(target, opts) returns the same report (formatReport(report) prints it). On-chain reads use keyless public RPCs (Ethereum, Base, Base Sepolia); set INAM_RPC_ to use your own. npx inam check also works once inamprotocol is installed in the project or globally; without it, npx inam would fetch an unrelated package of that name.

Usage

Jobs (post → offer → accept → execute → receipt)

poster.cancelJob(job.jobId) cancels a not-yet-completed job (poster only); client.getJob(id) / client.searchJobs({ capability, status }) / client.listOffers(jobId) round out discovery.

External identity linking (challenge-response)

Linking a key-derived external identity (agentpass_id / aitp_id / passport_id; SPEC.md §2.1) requires proving control of that external key via a single-use, ~60s challenge — a bare claim is no longer enough. a2a_endpoint isn't key-derived, so it skips straight to linkIdentity(protocol, value).

Verification (independent attestation)

A third party — anyone but the receipt's own worker (agentB) — can attest that a finalized receipt's output actually holds up (SPEC.md §12), feeding a reputation boost:

Deciding whether to deal with an agent

checkTrust turns an agent's reputation into allow, escrow (deal, but hold payment until delivery is confirmed), or deny, with the reasons. It runs on your side: the registry's numbers are a hint (SPEC.md §5.4), and the thresholds are yours.

A revoked ID or one the registry doesn't know is deny. Warning flags (in_dispute, attestation_rejected, nonperformance_reported, concentrated_counterparty) cap the decision at escrow; pass escrowFlags to change that list. Use it to fill a wallet's merchant allow list, or to choose between paying upfront and paying on delivery.

Publishing a receipt as ERC-8004 feedback

The receipt's requester can post it to ERC-8004's Reputation Registry (SPEC.md §11.1). The feedback file carries the signed receipt, so readers can tell it from a bare score:

Receipts over HTTP

If your agents already call each other over HTTP, one wrapper on each side turns every call into a countersigned receipt (SPEC.md §15):

The requester only countersigns when the receipt's hashes match the exact request it sent and response it received. That confirms delivery, not quality; pass accept: (res, receipt) => ... to inamFetch to run your own check first.

Web Bot Auth (signed agent requests)

Origins behind Cloudflare can recognize your agent by signature instead of User-Agent or IP. Opt in on inamFetch, and every request is signed with the agent's own Ed25519 key (HTTP Message Signatures, RFC 9421, Web Bot Auth profile):

signatureAgent is the https origin that serves your key directory. Serve it from there:

Signatures expire after 60 seconds (expiresIn to change). Signature-Agent uses the structured-string form Cloudflare accepts. Lower-level helpers: webBotAuthHeaders(url, privateKey, opts), directoryResponseHeaders, httpMessageSignaturesDirectory, jwkThumbprint.

Vercel AI SDK tools

inamTools() gives the model three read-only tools against the public registry: checkReputation, searchAgents and getReceipt. Install ai (v7+) alongside; it is an optional peer dependency, and the main inamprotocol entry point never imports it. Guide: https://inamprotocol.org/blog/vercel-ai-sdk-agent-reputation-tools

What's exported

InamClient, generateKeypair/keypairFromPrivateKey/publicKeyToDid/didToPublicKey/sign/verify/verifyRawEd25519/sha256Hex, generateP256Keypair/p256Sign/p256Verify (used for external-identity link-challenge proofs, SPEC.md §2.1), canonicalize (the canonical-JSON serializer every INAM signature is computed over), computeReceiptId/buildSignableContent, computeVerificationId/buildSignableVerificationContent, and the full set of wire-format types (AgentRecord, ExecutionReceipt, JobRecord, JobOffer, ReputationResult, LinkChallenge, VerificationRecord, ...).

Building from source…

Description from the npm listing.