KeelenAnswer — what it costs
What does it cost to run an AI coding agent on a real project?
Updated 21 August 2026
The bill has two halves: what the tool charges you, and what the model consumes. The second half is the one that surprises people, and it is driven by input tokens rather than by how much code gets written, because an agent re-reads the same files on every step of a task. We measured that. Across 12 controlled runs of the same specification, Claude Code, OpenAI Codex, Moonshot Kimi each shipped the same median of 3 merged pull requests, while their input-token use differed by about 4.7 times between the heaviest and the lightest. So the engine you choose can change the model half of the bill several times over without changing what gets delivered. Keelen charges a flat monthly fee (Indie $29, Operator $79, Agency $299) and your token usage bills to your own provider account at cost, with no markup added by us.
Those token figures measure our loop driving each engine, on small newly created projects, as medians with the sample size published beside each one. They are not a general claim about what any vendor's model costs in someone else's tool.
What to do, in order
Separate the platform fee from the model usage
Some products bundle model usage into one price and some run on credentials you connect. Bundled looks simpler and hides the variable. Connected looks like two bills and lets you see, and change, the part that actually moves.
Meter input tokens, not output
Output is the code, and it is small. Input is the context: the files read, re-read, and re-sent on every step. A modest feature can involve millions of input tokens and a few thousand output tokens, so an estimate built on lines of code will be wrong by an order of magnitude.
Price the unit you actually care about
Cost per merged pull request is the honest unit. Cost per run rewards a tool that gives up early, and cost per token rewards a tool that does less. Measure what reached your main branch and worked.
Check what your existing subscription already covers
If you already pay for a coding plan, running an agent on those credentials consumes an allowance you have bought rather than adding a metered bill. A heavier engine then costs you headroom instead of money, which is a different constraint.
Cap the autonomy you cannot supervise
The expensive failure is not an expensive task, it is a loop that retries a task it will never finish. Look for per-task retry budgets, a daily ceiling, and a project that pauses and asks rather than burning quota against a dead end.
Why input tokens dominate
An agent working on a real repository spends most of its budget reading. It opens files to find the relevant code, re-reads them as the task develops, and re-sends that context on every step of the conversation. The code it writes is a small fraction of the traffic. That is why two tools can produce the same change for very different money, and why the size of the diff predicts almost nothing about the cost.
What we measured, and the honest scope of it
12 controlled runs between 2026-07-20 and 2026-08-10, each giving every engine the same freshly written specification, the same starter repository, and the same merge gates. The projects were Python libraries and web apps, small and newly created. Every figure is a median with its sample size published beside it, and the sample is small enough that all of it is directional. We do not rank the engines on quality, because twelve runs cannot support that claim about anybody's product.
Whose account pays for the tokens
Keelen runs on credentials you connect: Anthropic Claude (a Claude Code subscription login or an API key, no Max plan required), OpenAI Codex, Zhipu GLM, or Moonshot Kimi. Token usage bills to your own provider account at cost. Keelen charges a flat monthly tier and never marks tokens up.
What stops a run from burning money on a dead end
Failed runs are classified into 30 kinds. Infrastructure blips and provider limits retry on a budget and never count against your work. A real dead end becomes one plain-English card in your Needs-you queue with a recommended action, and the project pauses rather than burning quota.
When Keelen is not the answer
- You want a single all-in bill with model usage included. Keelen does not resell tokens, so you will always see a provider bill or a provider allowance alongside the subscription.
- You have a handful of small tasks a month. A chat subscription you already own is cheaper than any continuous system, and honestly so.
- You need a firm quote before you start. Usage depends on your codebase and your work, so any number quoted in advance, here or anywhere else, is an estimate.
Hand the work to a loop
Connect a repository, write what you want in plain language, and review the tested pull requests that come back.
FAQ
What does it cost to run an AI coding agent every month?
Budget the platform fee plus your own model usage. The platform fee on Keelen starts at $29 a month. Model usage depends on your codebase and how much work you queue, and it bills to your own provider account at cost. The engine you pick moves that second number more than anything else you control.
Why is the token bill so much larger than the code produced?
Because the cost is dominated by reading rather than writing. An agent re-reads and re-sends the same context on every step of a task, so a change of a few hundred lines can involve millions of input tokens. Output tokens, the actual code, are a small part of the total.
Is a cheaper engine worse?
Not in what we measured. Every engine in our benchmark shipped the same median of 3 merged pull requests from the same specification, and the differences in speed and iteration count were smaller than the variation between individual runs. The clear difference was consumption, not delivery.
Can I use a coding subscription I already pay for?
Keelen runs on credentials you connect rather than on tokens we resell, and it supports Anthropic Claude, OpenAI Codex, Zhipu GLM, and Moonshot Kimi. What your particular plan permits is set by that provider's own terms, so check them before you rely on it.