Turn your AI chat accounts into an API — and chat with all of them at once

• A subscription and an API are two separate purchases

  A monthly plan buys you a web interface. It does not buy you programmatic access. The moment your code needs an answer, you are on a metered bill with no relationship to the flat fee you are already paying.

• Comparing models means doing everything twice

  The reason you keep three accounts is that they are good at different things. So you paste the same prompt into three tabs, wait three times, and hold three answers in your head to compare them. Every time.

• Every provider is its own integration

  A different SDK, a different auth scheme, a different response shape, another key in another secret store. Trying a second model to compare quality becomes a day of work before it becomes a comparison.

• Keys leak, and a leaked key spends your money

  A provider key in a repository, a log line, or a client bundle is someone else’s usage on your card. The safest key is the one that was never issued.

All four have the same root: the access you already have is stuck behind a browser tab, one tab at a time. Revoye is what happens when it is not.

1. 1

   It is queued before it is anything else

   The moment Revoye accepts your request, the job is written down and durable. If your connection drops, the job keeps running and the result waits for you at GET /v1/completions/{id} — or is delivered to your webhook. A client timeout loses nothing.

2. 2

   Your device is checked, not assumed

   Before anything is dispatched, the router checks whether one of your paired devices is actually connected. If none is, the work stays in your queue. When your machine comes back, the queue drains into it. A laptop that was closed overnight is a delay, not an error.

3. 3

   An eligible agent is chosen

   A candidate agent has to clear every one of these: enabled in your dashboard, enabled in your extension, idle, its provider enabled, its device connected, not already failed on this job, and under your agent, provider and global hourly limits. Whatever survives, your rotation strategy picks from. An agent busy with a chat prompt is skipped, not interrupted.

4. 4

   The prompt is typed, and the answer is read

   The agent opens the provider’s page, enters the prompt, and waits for generation to finish — not for a fixed timer, but for the page to actually stop producing. Then it reads the answer back.

5. 5

   A stuck agent does not become a stuck request

   Every attempt has a timeout. When one is hit, the job is taken off that agent and given to the next eligible one, which never includes an agent that already failed this job. The whole job has its own deadline covering all attempts, so retries cannot run forever.

6. 6

   The answer comes back the way it came

   The response returns through the extension, the desktop app and the router to your waiting HTTP request, or to your callback_url. Along with it: which provider answered, which agent, how many attempts it took, how long it queued, and how long it ran.

Nothing is authoritative in two places. Your job state lives in the database, your agents exist in your extension, and your provider sessions never leave your browser.

• Six answers to one question

  Send a prompt to every account you have and read the replies together — in a thread or in columns. The model comparison you were doing across six tabs, done once, in one place, in the time the slowest one takes.

• One integration for six providers

  One base URL, one bearer token, one request shape. Switching from ChatGPT to Claude is a string in a field, not a new SDK, a new auth scheme and a new response parser.

• No key to leak

  Revoye holds no provider credential, so there is no provider credential to steal from Revoye. Your Revoye key can be scoped down to write-only or read-only, rotated, and revoked from the dashboard without touching any provider account.

• Work that survives being offline

  Your queue is durable. Close the laptop mid-run and the job is still there when it wakes up. For anything overnight or unattended, this is the difference between a pipeline and a babysitting job.

• It plugs into what you already wrote

  Point an OpenAI SDK at Revoye by changing base_url. revoye/auto lets your routing policy pick the provider; revoye/chatgpt pins one. Your model picker keeps working, because /v1/models answers in the shape it expects.

• One place to see everything

  Every device, every provider, every agent and its live status, your queue depth, and the history of what ran, on which agent, and how long it took. Not six dashboards on six provider sites.

curl https://revoyeapi.degird.com/v1/completions \
  -H "Authorization: Bearer $REVOYE_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
    "prompt": "Summarise the CAP theorem in three sentences.",
    "provider": "chatgpt",
    "wait": true
  }'

{
  "id": "job_01JAY7Q2K8XYZ",
  "status": "succeeded",
  "response": "The CAP theorem states that…",
  "provider": "chatgpt",
  "agent_id": "agt_01JAY7…",
  "attempts": 1,
  "queue_ms": 240,
  "run_ms": 18432
}

curl https://revoyeapi.degird.com/v1/completions \
  -H "Authorization: Bearer $REVOYE_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Write release notes for the attached changelog.",
    "wait": false,
    "callback_url": "https://your-app.example.com/hooks/revoye"
  }'

{ "id": "job_01JAY7Q2K8XYZ", "status": "queued", "created_at": "2026-08-18T09:14:02Z" }

from openai import OpenAI

client = OpenAI(base_url="https://revoyeapi.degird.com/v1", api_key=REVOYE_KEY)

answer = client.chat.completions.create(
    model="revoye/chatgpt",          # or revoye/auto to let the router choose
    messages=[{"role": "user", "content": "Summarise the CAP theorem."}],
)

Read the full API reference →·Quickstart·Errors

Integrating with an AI coding agent? Download the Node.js integration skill — the whole public API in one Markdown file: authentication, every endpoint, the error table, retries, webhooks and a reusable client. Hand it to the agent and it has everything it needs in a single pass.

• Your provider sessions never leave your browser

  Revoye has no ChatGPT password, no Claude session cookie, no Gemini token. The agent works inside the browser where you are already signed in. There is no provider credential in our database because there is no step in the design that would put one there.

• Your device key never leaves your machine

  At pairing, Revoye Desk generates a keypair and stores the private half in your operating system’s keychain — Credential Manager on Windows, Keychain on macOS. Revoye receives only the public half. Every message from your device is signed with a key we do not have and could not disclose.

• The extension has no credential at all

  It cannot reach a Revoye server, and it holds no key, token or account identity. Its only outbound channel is a local bridge to Revoye Desk, which the operating system pins to exactly one extension ID. An attacker who reads your browser profile finds nothing of Revoye’s to use. It is also why the chat half works with no account: there is nothing for it to log in to.

• Your Revoye key is scoped, revocable, and shown once

  API keys carry scopes: write-only for a key embedded in a client, read-only for a monitor. We store a hash, never the key, so no endpoint exists that could show it again. An API key can never perform an account action — it cannot create another key, revoke a device, or read your audit log. A leaked key cannot become account access.

• We automate your session; we do not disguise it

  Revoye types at human speed, paces requests with the hourly limits you set, and behaves like a foreground tab so a background one still works. What it deliberately does not do: no user-agent or fingerprint spoofing, and no attempt to defeat a provider’s bot challenge. Revoye automates a session you are legitimately signed into, at human pace. That is the line, and it is why per-agent hourly limits are a real product feature rather than an afterthought.

API prompts and responses are stored so your job history works, and they are deleted on a retention schedule. If you need them gone sooner, delete the job. Chat conversations are saved in your own browser and go nowhere else.