Products / AI
Hoq Intelligence
One API key for many models, billed per token from a prepaid balance.
What it does
Hoq Intelligence is an AI gateway at ai.hoq.app. It accepts the OpenAI Chat Completions format and the Anthropic Messages format, so you point an existing SDK or tool at it by changing the base URL and the key.
Every request is authorized against your prepaid credit balance before anything goes upstream. If the balance is empty the gateway answers 402 with a top-up link, and if the ledger cannot be reached it answers 503. Usage is metered in raw input and output tokens and priced in the ledger, so each call on your dashboard shows the model, the tokens, the cache hits and the cost.
You can ask for a role instead of a model name. hoq/fast, hoq/default, hoq/reasoning and the other roles resolve to a priced model, and you can repoint a role for your account without changing any client code. Image, video and 3D generations run as async jobs, and a generation that fails is not charged.
In numbers
7named model roles you can repoint without a code change
9decimal places on every charge
2wire formats accepted: OpenAI and Anthropic
60tags per account for grouping spend by API key
Specifications
| Wire formats | POST /v1/chat/completions (OpenAI, streamed over SSE), POST /v1/messages (Anthropic, streamed over SSE), POST /v1/embeddings and GET /v1/models. |
|---|---|
| Authentication | Hoq ID API keys with the scope ai:infer, sent as Authorization: Bearer or x-api-key. |
| Billing model | Prepaid credits. A request with no balance behind it is refused with HTTP 402 before any upstream call. |
| Metering units | Input tokens and output tokens per request, with cache reads and cache writes reported separately. Generation jobs are metered in job units: video per second of requested duration, 3D per request. |
| Charge precision | Credit amounts are stored to 9 decimal places, and a line under one cent shows its exact value instead of 0.00. |
| Early disconnects | Metering runs on every exit path, and each request meters under its own idempotency key, so a retried meter call debits once. |
| Failed generations | Async generation jobs are charged only when they complete. Failed, cancelled and timed-out jobs cost nothing. |
| Prompt caching | On the Anthropic format, the gateway adds up to two cache breakpoints when the client sent none. |
| Model roles | hoq/fast, hoq/default, hoq/reasoning, hoq/vision, hoq/image, hoq/background and hoq/embed, each resolving to a priced model you can repoint per account. |
| Who pays | An organization can pay for its members' AI, with an optional daily cap per member. |
| Request logs | The access log records status and latency per request. It keeps no request bodies, no keys and no query strings. |
Features
15 features, all shipped. Listed from the product's code and documentation; each row shows the day it was last checked.
| OpenAI format | Streamed chat completions in the OpenAI format, with usage reporting switched on for you. | |
| Anthropic format | Streamed Messages API calls, so Anthropic SDKs and tools work with a new base URL. | |
| Embeddings | Text embeddings through the same key, priced on input tokens only. | |
| Model list | GET /v1/models answers in the OpenAI list shape, including your role table. | |
| Async generation jobs | Submit image, video or 3D generations to /v1/jobs and poll status and result URLs. |
| Model roles | Ask for a role such as hoq/reasoning and the gateway picks the priced model behind it. | |
| Role preferences | Repoint a role for your account, with the price shown beside each candidate. | |
| Pooled models | Some models are served from several subscription accounts, and a request moves to the next account when one runs out or fails before sending any bytes. |
| Prepaid balance | Buy credits by card at id.hoq.app/billing; the balance goes up when the payment provider confirms the charge. | |
| Fail closed | No key returns 401, no funds return 402 with a top-up link, an unreachable ledger returns 503. | |
| Organization billing | An organization can pay for members' AI and set a daily cap per member. |
| Usage view | Balance, daily spend, spend by model and per-request lines with tokens, cache hits and cost. | |
| API keys and tags | Mint keys and tag them many-to-many to see spend by client or team. | |
| Playground | Try a model in the browser against your own key, with the running cost under the composer. | |
| Refusals on the record | Refused requests and failed jobs appear in your usage with the reason, at zero cost. |
Works with
- GrimasteGrimaste agents think through Intelligence and their AI usage bills to your balance.
- Hoq AgentsAgents can run on Intelligence models instead of a separate provider account.
Your data
| Where it is kept | Usage records and the balance ledger are kept in your Hoq ID account. Prompts and completions pass through to the model provider and are not written to the gateway's access log. |
|---|---|
| Getting it out | Not published. |
| If you leave | Not published. |
APIs and standards
- OpenAI-compatible API: https://ai.hoq.app/v1/chat/completions, /v1/embeddings, /v1/models
- Anthropic-compatible API: https://ai.hoq.app/v1/messages
- Async jobs: POST https://ai.hoq.app/v1/jobs/{endpoint}
- Web dashboard at intelligence.hoq.app
- Standard: OpenAI Chat Completions API
- Standard: Anthropic Messages API
- Standard: Server-Sent Events
Compared
| Checked 2026-11-01 | Hoq Intelligence | OpenRouter | Anthropic API (direct) |
|---|---|---|---|
| Request formats | OpenAI Chat Completions and Anthropic Messages | OpenAI-compatible: Chat Completions and Responses1 | Anthropic Messages2 |
| Model catalogue | 7 named roles plus the priced models behind them; full list shown after sign-in | 600+ models from 80+ providers3 | Claude models only2 |
| Fee on top of model price | Markup built into each per-token price, not published as a percentage | 5.5% on credit purchases (Standard plan)3 | not published2 |
| Spend control | Prepaid balance; requests refused with 402 at zero; per-member daily cap for organizations | Budgets and per-key credit caps (Standard and above)3 | Monthly spend cap per tier: $500 Start, $1,000 Build, $200,000 Scale; your own lower limit2 |
| Teams | Organization pays for members' AI | 5 workspaces on Standard, 1,000 on Business3 | Workspaces with their own spend and rate limits2 |
| Zero data retention routing | not published | Available on every plan3 | not published2 |
| SSO | not published | Enterprise only3 | not published2 |