Run background coding agents on the subscription you already pay for

Updated 2026-09-02

A coding-agent CLI login is bound to the machine it was authenticated on, which is why a cloud agent asks for an API key instead of your subscription. Denly runs the CLI where that login already lives, so background work bills to the plan you pay for every month rather than to a second metered account.

Denly is a control plane for coding agents (Claude Code, Codex, Cursor, Grok Build, Antigravity) that run on your own machines and your own subscription: you describe a task, the agent plans it, you approve, it builds locally and opens a pull request. Inference is billed directly by your agent vendor; Denly charges for the control plane and nothing else.

Which hosted agents can use your plan, and which cannot

Subscription logins for agent CLIs are issued to the machine that authenticated, not as a portable token you can hand to somebody else's server. That splits hosted agents in two. A first-party cloud agent meters against its own vendor's plan, because the vendor holds the account: OpenAI's Codex cloud tasks draw on your ChatGPT plan. A third-party hosted service cannot reuse the CLI session on your machine, so it asks for an API key it can meter or resells inference at its own margin.

Running the CLI on your own hardware is the case where any vendor's subscription applies, including the vendors with no cloud agent of their own.

The five runtimes, and what each reports

Runtime	How it authenticates	Reports quota to Denly
Claude Code	Its own login on the runner machine, or a stored API key	Yes
Antigravity	Its own login on the runner machine	Yes
OpenAI Codex	Its own login on the runner machine	No quota published
Cursor	Its own login on the runner machine	No quota published
Grok Build	Its own login on the runner machine	No quota published

Usage shows the quota a provider reports plus the tokens Denly counted, separated into task work and Chat. Two of the five publish a quota; the other three do not, and Denly shows what it has rather than an estimate.

Set the runtime, per project or per workspace

Install and authenticate the agent CLI on the runner machine. denly authmode reports how each installed CLI is authenticated there.
Pick the workspace default runtime and model, then override it per project where a project needs a different one.
To use an API key instead of a subscription, denly set-key stores it in the machine's own secret store – not in a file, and not on Denly's servers. denly clear-key removes it.
A task can also name its runtime at creation, and a failed run can be retried fresh on a different one.

What Denly costs on top

The control plane is what you pay Denly for, and the agent vendor bills your inference separately:

Plan	Price	What it covers
Free	$0	Unlimited projects, 2 machines, one task at a time, 250 MB, 1 person
Pro	$10 per month	One person: unlimited workspaces, projects and machines, parallel tasks, 5 GB pooled
Teams	$12 per seat per month	Everything in Pro for a team; 10 GB + 2 GB per seat; 2-seat minimum, 14-day trial

Pro and Teams can be billed monthly or yearly; the interval is chosen at checkout. Prices here are the ones published on the pricing section of the site.

The trade-off

Using your own subscription and running in the cloud around the clock cannot both be true, because the login lives on the machine. Work runs while your machine is awake and online, and queued work waits for the runner to wake.