rivr

A read-only Amazon Shopping CLI built for AI agents.

Amazon's Product Advertising API was retired in 2026 and no agent-safe replacement existed — so rivr is one. Search products, fetch detail/offers/reviews/variations through one normalized schema over a pluggable backend (SerpApi, Rainforest, the official Amazon Creators API, or keyless scraping), with the Agent CLI Guidelines baked in: read-only by default, structured errors + stable exit codes, bounded JSON, prompt-injection fencing, and stdin/keyring secrets.

Why rivr

• Read-only by design. No cart, no checkout, no mutations. Every result is structured data plus a canonical /dp/<ASIN> deep link — the hand-off point where a human buys in their browser. There's no destructive action for an agent to misfire.
• Built for agents, not just humans. --json everywhere, schema --json, an embedded agent guide, {error,code,remediation,retryAfterSeconds} errors, a stable exit-code table, and --limit/--select so responses stay within an agent's context budget.
• Prompt-injection fenced. Titles, descriptions, and review text are attacker-controllable; rivr wraps them ‹untrusted›…‹/untrusted› by default.
• Pluggable backend, one schema. Swap SerpApi ↔ Rainforest ↔ Creators ↔ scrape without relearning fields. rivr provider list --json describes each backend's capability/cost/risk.

Install

# Homebrew (macOS / Linux) — recommended
brew install rnwolfe/tap/rivr

# Go (any platform; static, no CGO)
go install github.com/rnwolfe/rivr/cmd/rivr@latest

# Shell script (macOS / Linux) — downloads a release binary, verifies its SHA-256, installs to ~/.local/bin
curl -fsSL https://rivr.sh/install.sh | sh

Or grab a prebuilt binary (linux/macOS/windows, amd64/arm64) from Releases — each ships with checksums, an SBOM, and build-provenance attestation.

Quickstart (first success in <60s, no account)

# the offline 'stub' backend returns fixtures — no key, no network
RIVR_PROVIDER=stub rivr search "usb-c cable" --json
RIVR_PROVIDER=stub rivr provider list --json --select name,keyless,hostedSafe,reviewsScope
rivr schema | jq '{read_only: .safety.read_only, exit_codes}'

Then point it at real data — the lowest-friction path is SerpApi's free tier:

# one-time, key read from stdin (never argv)
printf %s "$SERPAPI_KEY" | rivr auth login --provider serpapi
rivr search "mechanical keyboard" --prime --min-rating 4 --json
rivr item get B0XXXXXXX --detailed --json

No key at all? On a home/residential connection you can use the keyless scraper (opt-in, Amazon-ToS caveats apply — see the backend matrix): RIVR_SCRAPE_ENABLE=1 rivr --provider scrape search "usb-c cable" --json.

Providers & authentication

rivr normalizes one schema across backends; pick with --provider (or RIVR_PROVIDER). Secrets are read from stdin (never argv) and stored in the OS keyring, with a 0600 file fallback (RIVR_KEYRING=file forces it on headless boxes).

Provider	Get a credential	Notes
serpapi (default)	sign up → serpapi.com/manage-api-key → printf %s "$KEY" | rivr auth login --provider serpapi	renewing free tier (~250/mo); reviews are a page sample
rainforest	Traject Data dashboard → printf %s "$KEY" | rivr auth login --provider rainforest	full paginated reviews + real offers
creators (official)	Associates Central → Tools → Creators API → printf '%s\n%s' "$ID" "$SECRET" | rivr auth login --provider creators	needs an approved Associate w/ ≥10 sales/30d, else ASSOCIATE_NOT_ELIGIBLE; no review text
scrape (keyless)	RIVR_SCRAPE_ENABLE=1 (opt-in)	home/residential only — modest throttled requests; ToS risk + fragile. Don't run from cloud/hosted IPs (→ use creators).

Capability / cost / risk at a glance

backend	keyless	hosted-safe	reviews	cost	headline risk
serpapi (default)	✗	✓	sample	free tier + paid	paid beyond free tier; sample reviews
rainforest	✗	✓	full	paid (~$25/mo+)	paid per request
creators (official)	✗	✓	none	free*	*needs eligible Associate; no review text
scrape	✓	✗ residential	none	free	Amazon ToS + blocking; no review text; home use only
stub	✓	✓	fake	free	not real data (testing)

Full matrix + "which should I use?" → docs/backends.md. Machine-readable for agents: rivr provider list --json.

rivr auth status --json tests the active provider; rivr auth logout removes local creds only (revoke at the provider to invalidate server-side); rivr doctor --json runs full diagnostics. A missing credential returns AUTH_REQUIRED (exit 4) naming the login command.

Commands

rivr search <query>          # keyword search (--category --prime --min-rating --min/max-price --sort)
rivr item get <ASIN>...      # product detail (--detailed adds bullets/specs)
rivr item offers <ASIN>      # live offers / buybox / availability
rivr item compare <ASIN>...  # side-by-side + best-of summary (cheapest/highest-rated/most-reviewed)
rivr reviews <ASIN>          # customer reviews (scope: full | sample)
rivr variations <ASIN>       # size/color/style variations
rivr browse <node-id>        # category (browse-node) tree  [creators only]
rivr provider list           # backends + capability/cost/risk
rivr auth login|status|logout|refresh
rivr doctor                  # diagnostics
rivr schema                  # machine-readable command tree, flags, exit codes, safety
rivr agent                   # print the embedded agent guide (SKILL.md)
rivr version [--check]       # installed version; --check pulls GitHub for the latest release

rivr never auto-updates. On the human path it prints a once-a-day upgrade hint to stderr (silent for agents; disable with RIVR_NO_UPDATE_CHECK=1).

Global flags: --json/--format json|plain|tsv, --select a,b.c, --limit N, --detailed, --provider, --associate-tag/--no-associate-tag, --no-wrap-untrusted, --no-input.

Exit codes

0 ok · 2 usage · 3 empty results · 4 auth required · 5 not found · 6 permission / ASSOCIATE_NOT_ELIGIBLE · 7 rate limited / blocked · 8 retryable · 9 upstream / schema drift · 10 config · 11 unsupported by provider · 13 input required. Full table: rivr schema.

Conformance

rivr conforms to the Agent CLI Guidelines v0.4.0 at the Full level. The contract version is machine-verifiable from the binary — rivr schema emits a conformance block, so an agent can confirm the standard it was built against without trusting this README:

{
  "conformance": {
    "level": "Full",
    "spec": "agent-cli-guidelines",
    "version": "0.4.0"
  }
}

Affiliate attribution (how rivr is funded)

Product deep links are tagged with an Amazon Associates ID by default — if a referred link leads to a purchase, Amazon pays the project a small referral fee, the buyer pays nothing extra. It's fully transparent and in your control: the active tag is in every url, in rivr doctor, and in rivr schema. Use your own with --associate-tag <id>, or disable with --no-associate-tag.

As an Amazon Associate the maintainer earns from qualifying purchases.

Safety

• Read-only. No cart/order/mutation surface; the deep link is the action boundary.
• Untrusted text fenced by default (--no-wrap-untrusted to disable).
• Secrets via stdin/env → OS keyring (0600 file fallback); never argv; redacted in output.

See SECURITY.md for the secret-handling threat model.

Contributing

See CONTRIBUTING.md and AGENTS.md. Build/test: go build ./... && go vet ./... && go test ./....

License

MIT © Ryan Wolfe