OptiQ Cloud

Boost runs a task with optiq-boost, the OptiQ Cloud model, in charge. In OptiQ Code the cloud model also gets your local model as a helper: it plans and makes the decisions, and hands the reading, searching and test runs to the model on your Mac. Any other client can pick optiq-boost as a model through optiq serve. You pay for the tokens it uses.

Connect a machine

Sign in once per machine. The browser approves the request; the token is written to your OptiQ config and stays on the machine.

$ optiq cloud login

optiq cloud status shows the account and the balance. optiq cloud logout revokes the token rather than only forgetting it, so a lost laptop cannot keep spending. A running optiq serve picks up a new login on the next request, without a restart.

Boost in OptiQ Code

Type /boost before a hard task. From the next task until you type it again, the cloud model leads and your local model is its local tool. The status bar shows when Boost is on.

/boost
add retry with backoff to the HTTP client and make the tests pass

To have it on from the start, set boost_mode to lead in your OptiQ Code config or with OPTIQ_BOOST_MODE=lead. The default is off.

The local tool runs your local model in its own fresh context with the same repository and tools. The cloud model uses it for work it can describe in a sentence: read these files and answer a question, run the tests and report only the failures, make an edit it has fully specified. Only the short answer comes back to the cloud model, so long files and command output stay on your Mac. By default the helper is the model your OptiQ Code would otherwise use; set local_tool_base_url and local_tool_model to use a different one.

At the end of a run the summary shows what the task cost and what is left:

tokens: 61,240 in / 1,830 out  ·  48,112 cached (79% of in)  ·  OptiQ Cloud $0.0062 · $1.99 left
local tool (Qwen3.6-35B-A3B): 2 calls, 14 turns, 210,512 in / 4,230 out

Boost in other clients

Once you are signed in, optiq serve and the Lab list optiq-boost next to your local model. A request that names it is sent to OptiQ Cloud with your login, and the answer comes back in the client's own format. Every other model name is answered by your local model as before.

In these clients the cloud model works on its own, without the local helper, which only OptiQ Code provides. Only clients on the machine running optiq serve can use optiq-boost: optiq serve checks the shape of a token, not the token, so on --host 0.0.0.0 anyone on your network could otherwise spend your balance. See Integrations for each client's setup.

Pricing

You pay per token, from a dollar balance.

A long agent run re-sends its conversation on every call, and most of that is served from cache, which is why the cached rate matters. On our 30-task benchmark the median task cost $0.056 and the most expensive $0.33.

Sign-up includes $2 of credit, good for 14 days. Credit is bought $20 at a time, and is not a subscription. Auto top-up is optional, with a monthly cap you set. If you had Boosts from before per-token billing, each one left was converted to $0.10 of credit.

Seeing what you spent

The dashboard shows your balance, this month's spend, and every recent task with its calls, input, cached and output tokens, and cost. Calls are grouped into a task by the session id the client sends; OptiQ Code and Claude Code send one, and other clients' calls are listed one by one. It also lists the credit you added.

Limits

There are no hourly, daily or per-task caps. Your balance is the limit: a call is accepted while it is above zero and billed afterwards for what it used, so the last call can take it slightly below. When it is empty, OptiQ Code stops the run and says so, and other clients get the same message as an error. The client's own limits still apply, such as OptiQ Code's turn limit and wall-clock limit.

What leaves your machine

The conversation of a task that runs on optiq-boost: your request, the cloud model's replies, and the tool results it asks for, including the parts of files it reads. Tasks on your local model never reach the network, and neither do the jobs the cloud model hands to your local model, only their short answers. Your repository is not uploaded, indexed or scanned.

OptiQ Cloud does not store prompts or replies, and Boost only uses model providers that do not train on requests. What is kept is token counts, timings and cost, so the dashboard can show what a task cost.

When Boost cannot run

Each case comes back as a plain message in the client's own format:

Accounts, keys and billing

Signing in, connecting a machine, how your coding agent gets the key (it does not), API tokens and auto top-up are on Accounts and billing.