Redrob Console
One balance, andthe right model for Code.
Pay once, use the credit across every Redrob application, and see what each request cost.
Dashboard
Credit balance
$42.18
Top up
Spend, last 30 days
$11.97
across 5,249 requests
Auto
You are not choosing a model any more
Ask for auto. Redrob reads each request, sends it to a model that can do it, and keeps getting better without a deploy.
Cheaper to serve is cheaper to buy
From $0.02 per million input tokens, $0.12 per million output: what the answering model cost us, plus 5%. A cheaper model is a cheaper request.
Send an image, not a description of one
Put an image or an audio clip in the message. Routing only considers models that can read what you sent.
A bad hour is routed around
A degraded vendor becomes another vendor’s request rather than an error your users read. Nothing has to be redeployed for that to happen.
Routing is a default, not a lock
Pin a model when you have tested against one, and every request is logged with its tokens, latency and outcome whichever model served it.
One line changes. The rest of the stack does not.
The completion API speaks OpenAI’s chat completions protocol. Point the official SDK at Redrob with a Redrob key and the code you already shipped keeps working - what changes is which model serves each request and what it costs.
from openai import OpenAI
client = OpenAI(
api_key=os.environ["REDROB_API_KEY"],
base_url="https://console.redrob.ai/api/backend/v1",
)
answer = client.chat.completions.create(
model="auto",
messages=[{"role": "user", "content": "이 계약서 요약해줘"}],
)Three applications finish the work together
Code works in the terminal you already have. Design turns a brief into editable visuals, and Cowork does the work on the files on your own machine.
- CodeCoding help in your terminal, in the language you write in. Installs with one command on macOS, Linux and Windows.macOS, Linux, WindowsWhat it isRC | Fix broken tests
› Fix broken tests
The retry helper drops the last attempt when the backoff overflows. Fixing the guard and adding the case.
- ✓Readsrc/queue/retry.ts
- ✓Grepbackoff
- ✓Editsrc/queue/retry.ts
- ✓Bashyarn test queue
4 passed, 0 failed.
▸ Fix a TODO in the codebase
~/work/checkout3 MCP2 LSP1 Permission - CoworkA desktop workspace for repeatable team tasks, shared prompts and connected tools. Stable Windows, macOS and Linux builds are available now.Windows, macOS, LinuxWhat it isNew session
What do you need done?
Describe it in plain language
Describe your task...
Run task- Research and write a brief
- Clean up a spreadsheet
- Draft a document
- Automate a web task
- DesignLayout and image work driven by a description rather than a toolbar. A signed Windows installer and Linux packages you can download now.Windows, LinuxWhat it isUntitledMoveFrameRectangleEllipsePenText
Three steps, then you are using it
No plan to compare and no setup to get through. The first payment is the only thing between signing up and using Redrob.
Sign up
An email and a password. You become the admin of a new workspace, and you can invite the rest of your team whenever you want to.
Pay with a wallet tap
Pick an amount from $5 and pay. Stripe shows the methods available where you are, which can include Apple Pay, Google Pay, Kakao Pay and Naver Pay as well as a card, so there is usually no form to fill in.
Connect Redrob in your app
Your credit is spent from whichever Redrob application you open, and Console shows what each one cost. The guide walks through it with no code in it.
One inference layer, across models and languages
Model providers, language handling, metering, and access control usually live in separate systems. Redrob puts them behind one API so your application stays stable while the infrastructure underneath changes.
- Answers in the language your users type
- Send the prompt as you received it and the answer comes back in the same language. Translating is a separate call you make when you want it, so a multilingual product does not quietly cost twice as much.
- Metering you can reconcile
- Every request is logged with its tokens, latency, and cost, and the usage figures are aggregated from those same rows. Nothing in a dashboard that you cannot trace to a request.
- Built for a team, not a script
- Invite teammates with viewer, developer, or admin roles. Give each API key its own system prompt so staging and production behave differently without a code change.
15 languages, the same endpoints
English, Korean and thirteen South Asian languages share the same completion, streaming and translation APIs. Script detection is local, instant, and deterministic.
- English
- हिन्दीHindi
- বাংলাBengali
- தமிழ்Tamil
- తెలుగుTelugu
- मराठीMarathi
- ગુજરાતીGujarati
- ಕನ್ನಡKannada
- മലയാളംMalayalam
- ਪੰਜਾਬੀPunjabi
- اردوUrdu
- ଓଡ଼ିଆOdia
- অসমীয়াAssamese
- नेपालीNepali
- 한국어Korean
Every call is priced where you can see it
One row per request, with the key that made it, the tokens it used and what it cost. Your prompts are not kept.
Logs
| Time | Key | Status | Tokens | Latency |
|---|---|---|---|---|
| 09:41:02 | sk-live-…4f2a | success | 1,284 | 412ms |
| 09:40:57 | sk-live-…4f2a | success | 866 | 298ms |
| 09:40:51 | sk-ci-…9b10 | success | 2,140 | 731ms |
| 09:40:44 | sk-live-…4f2a | success | 512 | 204ms |
Prepaid credit, not a plan
No monthly fee, no seats, and nothing to cancel. Buy credit from $5, spend it as you use Redrob, and top up when you want to.
Your first payment
$5
minimum, one off
Roughly 58,000 conversations: a question with some context behind it and an answer of a few paragraphs. It is an estimate, and the console shows what you actually spent.
- No plan to choose
- No free trial, no welcome credit
- Card or wallet, through Stripe
Top up when you want
From $5
any amount, any time
One balance for the whole workspace, drawn on by every Redrob application and by the API. It does not expire, and there is nothing to renew.
- Spend it across every application
- Per-request cost in the log
- Runs out and requests stop
Or let it top itself up
+$0
no premium for the convenience
Set the balance to refill at and the amount to add, and an unattended integration keeps running without anyone watching the number.
- Same rate, no premium
- Off by default
- Changed or stopped in Console
What it cost us, plus 5%
A short question
500 in / 300 out
- mistralai/mistral-nemo
- $0.0000252,631/$1
- gpt-5.6-terra
- $0.00483207/$1
- claude-sonnet-5
- $0.00630158/$1
- claude-fable-5
- $0.02147/$1
A coding turn
8,000 in / 1,500 out
- mistralai/mistral-nemo
- $0.000214,830/$1
- gpt-5.6-terra
- $0.03628/$1
- claude-sonnet-5
- $0.04920/$1
- claude-fable-5
- $0.1636/$1
A long document
60,000 in / 2,000 out
- mistralai/mistral-nemo
- $0.00126793/$1
- gpt-5.6-terra
- $0.1516/$1
- claude-sonnet-5
- $0.2214/$1
- claude-fable-5
- $0.7351/$1
Every model is billed at what it costs us, plus 5% for the routing, metering and keys around it. The cost is published beside the price, so the cut is checkable rather than asserted. `auto` is billed the same way, at whatever model it routed the request to.
| Model | Input | Output |
|---|---|---|
| mistralai/mistral-nemotool calling · structured output | $0.0199 ← $0.0190 | $0.0315 ← $0.0300 |
| gpt-5.6-terratool calling · structured output · million-token contextaccepts images | $2.10 ← $2.00$4.20 above 272K | $12.60 ← $12.00$18.90 above 272K |
| claude-sonnet-5tool calling · structured output · adjustable reasoning (low, medium, high, xhigh, max) · priority tieraccepts images | $3.15 ← $3.00 | $15.75 ← $15.00 |
| claude-fable-5tool calling · structured output · adjustable reasoning (low, medium, high, xhigh, max) · priority tieraccepts images | $10.50 ← $10.00 | $52.50 ← $50.00 |
Usage
Daily cost (USD)
The things people ask first
No. Redrob is prepaid credit: you buy an amount from $5, it sits in your workspace as one balance, and requests draw it down as you make them. There is no subscription, no seat price and nothing to cancel.
Auto top-up is there if you would rather not think about it: set a threshold and an amount, and the balance refills itself.
No. A new workspace starts at zero and its first step is a payment. We would rather say that here than advertise a trial and take it away later.
Console: paying, the balance, per-request logs, usage by day and by key, team roles, the playground and translation. The API is open too, and Redrob Code installs from its own page with one command.
Cowork and Design have published builds you can download: a Linux desktop build and a signed Windows installer each. Both pages link the files and say what the file is.
No. It stays in the workspace until it is spent, and every Redrob application draws on the same balance rather than on its own.
If it runs out, requests are refused rather than billed on. There is no invoice arriving after the fact.
English, Korean and 13 South Asian languages share the same completion, streaming and translation endpoints. Send a prompt as you received it and the answer comes back in the same language.
This site and the console are in English and Korean, and the language control is in the header and the footer.
The log records what a call cost: the key that made it, the model that served it, token counts, latency and the request id. The usage figures are aggregated from those same rows, which is why the two always agree.
Yes, and none of it goes through an application. The API speaks the OpenAI chat completions protocol, keys are issued and revoked in Console, and a request bills the same workspace balance.
Ask for a specific model when you have tested against one, or ask for auto and let requests route across vendors.
Start with $5.
Create a workspace, pay with a wallet tap, and the credit is there to use across every Redrob application. Nothing to choose, nothing to cancel, and nothing expires.