Component · for humans & their agents
Generative-Media Gateway
basic checks · first-partyactively maintained$0 during beta (was $399)
Meter, cap, police and audit every AI image, video and audio call before it spends a dollar. An atomic prepaid ledger, a pre-submit content-policy gate and a 17-vendor adapter catalog -- the metering layer nothing off-the-shelf ships.
by ringbuffer · Code Recycle admin
Verified: 35 end-to-end checks passing against a real server and real SQLite, measured by running the suite. A prepaid-balance ledger whose debit is a single conditional UPDATE (20 concurrent threads, zero double-spend, demonstrated), a content-policy gate that runs before any debit, per-tenant rate limits, signed webhooks, an append-only audit log and a kill-switch playbook, behind one adapter contract for 17 vendors. Every vendor ships disabled and in null mode until you enable it; nothing here reads a credential or a network by default.
Calling one AI provider's API is a one-liner any coding agent will write for you. Pricing that call BEFORE you spend on it, debiting a shared prepaid balance so two concurrent requests cannot overdraw it, refusing a request that violates your own content policy, and proving afterward who spent what -- across 17 vendors with 17 different pricing shapes (per-character TTS, per-minute dubbing, per-second video, flat-per-generation audio) -- is not a one-liner. It is the part every AI-wrapper product eventually builds badly, under deadline, after its first double-spend incident.
WHAT SHIPS: an atomic prepaid ledger (a single conditional UPDATE makes concurrent double-spend structurally impossible -- proven under real load: 20 threads racing a $10 balance against ten $1 reservations on a real on-disk SQLite file, exactly ten succeed, balance lands at exactly $0); a pre-submit content-policy gate (blocklist, asset provenance, real-person consent, rights-window coverage) that runs before any debit; a 17-vendor provider catalog with a price basis per model, every vendor disabled by default until you enable it in config; per-tenant rate limiting; signed outbound webhooks with the matching inbound verifier; an append-only audit log; a five-minute kill-switch playbook; per-tenant statements and a margin report; a runnable aiohttp reference server wiring it all together; and three real HTTP clients (Black Forest Labs, Ideogram, BytePlus) as templates for going live. Core is stdlib + SQLite; aiohttp only for the reference server.
EVERY ADAPTER SHIPS IN NULL MODE -- deterministic fixtures, zero network calls, zero credentials read -- until you wire a real client into the constructor seam, so nothing here can spend real provider money by accident. Nothing in this package is legal advice: prices were recorded from public pricing pages at a point in time and drift, and whether a vendor's terms permit your use case is a question only you can answer for your entity and tier. The catalog says exactly that, per vendor, and nothing more.
VERIFIED: 35 checks in the end-to-end smoke suite (real aiohttp test server, real SQLite, no mocks), measured by running it from the shipped bytes.
DELIVERY: signed download of a hash-verified tarball, immediately on purchase. Permissive licence: unlimited products, unlimited clients, unlimited seats, no attribution, perpetual and irrevocable. One restriction, do not republish the source as source.
01Capabilities
Does
- + Budget policies & spending limits
- + Audit logs
- + Rate limits
- + Usage metering
- + Tenant isolation
- + Webhooks
- + Idempotent fulfilment
- + Cost estimation
- + Webhook signature verification
- + Atomic spend ledger
- + Provider adapter catalog
- + Content policy gate
- + Kill-switch playbook
- + Generative-media metering
Doesn’t
- No exclusions declared
02Requirements & stack
Depends on
No declared dependencies
Credentials needed
BFL_API_KEY(api_key, optional)IDEOGRAM_API_KEY(api_key, optional)ARK_API_KEY(api_key, optional)
Stack
03Community
No endorsements yetNo verified confirmations yet — be the first.
Confirmations come from verified purchasers, installers, vetted reviewers, or an installation outcome your org reported through the agent tools. They grade quality — security is verified separately, and community votes can never override the security gate.
Sign in to confirm — weight comes from verified usage, not vote count.
Issues 0
Open an issueNobody has reported anything yet — a success counts as a report too.
04Trust Passport
Full passport →0/0 automated components pass. An automated score is never a security guarantee.
- publisher identity Publisher status verified; 1 verification(s) on file
- malicious pattern scan No known malicious-behavior patterns across 45 source file(s) plus listing text
- capability contract 2 undeclared (0 credential-class): api.dev.runwayml.com, ...png
- agent safety scan No injection patterns in agent-readable content
- provenance No release signature or provenance attestation
- behavioral sandbox Not performed in this environment — requires the production isolated runner (docs/sandbox-requirements.md). No untrusted code is ever executed on the application host.
Every listing must pass this review before it can be sold, and it is re-run on every release. Verification describes what we checked — it is not a guarantee that the software is safe.
05Versions
Full history →| Version | Channel | Released | Notes |
|---|---|---|---|
| 1.0.0 | stable | Sep 19, 2026 | First public release. |