Run your first cloud AI coding task

What is a cloud coding agent in Lispa?

Lispa gives an AI coding agent a Linux cloud machine to work on your GitHub or GitLab repository. Describe a task, let the agent inspect and change the code, then review the diff and test output in the browser.

You can keep working in your editor while a separate cloud task runs. Claude, Codex, Grok, Gemini, and open models are available through supported connections or Lispa credits. Switch models yourself or let routing choose among the models available to your workspace; you can turn routing off.

Start with one reproducible bug

Choose a problem with a clear failure and a checkable result. For example, a pagination endpoint that repeats the final item from page one at the start of page two is easier to evaluate than “improve the backend.”

1. Create an account and choose a plan, then connect your GitHub or GitLab repository.
2. Choose the branch to work from and describe the failure. Include reproduction steps, relevant files if you know them, and the test command your project uses.
3. Set the boundaries. For a first task, ask for a fix and a regression test, and explicitly ask the agent not to push or merge until you have reviewed it.

An example prompt

On [branch], page two of [endpoint] repeats the last item from page one.
Reproduce the failure and identify its cause.
Add a regression test that fails before the fix and passes afterward.
Keep the change limited to this bug. Run [test command].
Show the diff, test output, and any checks you could not run.
Do not push or merge.

Replace the brackets with your repository details. This is a suggested task, not a claim that the agent has fixed a real repository or passed its tests.

Review evidence before merging

The workspace exposes the conversation, terminal, files, diff, and browser. Use them to check the result rather than relying only on the agent’s summary.

• Reproduction: does the original test or reproduction show the reported failure?
• Scope: does the diff address the cause without unrelated changes?
• Checks: did the relevant tests run, and are skipped or failing checks explained?
• Behavior: where applicable, does the changed app work when you inspect it in the browser?
• Usage: check model use and estimated cost in the chat, and charges in Usage.

Ask for a follow-up if anything is missing. Agents can push or merge when instructed, so set explicit permissions for each task and automation. Human review remains your responsibility.

What does a first task cost?

Lite is $20 per organization per month, including 20 usage credits and unlimited members. One credit equals $1. Prices are in USD, excluding tax. Compare plans.

The subscription does not buy a fixed number of completed tasks. Usage depends on the models, tokens, machine size, runtime, and services used:

• Models on Lispa’s keys: provider model cost plus 10%.
• Your connected plans or API keys: no Lispa model-call charge; your provider’s own billing and usage limits still apply.
• Cloud machines: Standard costs 0.53 credits per running hour; larger sizes cost 1.05 or 2.10. Machines stop when idle.
• Routing and knowledge services: token charges at 20% of the session model’s rates, including when you bring your own plan.

For example, one hour on a Standard machine uses 0.53 credits for the machine alone. Model and service charges are additional. This is a cost calculation, not a task completion estimate.

Monthly plan credits reset. Extra credits cost $1 each with a $5 minimum and carry over; auto top-up is off by default. You can cancel in Billing and retain access through the paid month. Check the current pricing and usage details before subscribing.

Which coding tasks can you delegate?

Choose a task with a clear input and a result you can verify. Each workflow below includes a sample brief, expected review evidence, and a starter prompt.

• Feature development: provide acceptance criteria; review the implementation, tests, and diff.
• Bug fixing: provide reproduction steps or a failing test; check the cause, patch, and regression coverage.
• Code review: provide a branch or pull request; inspect findings tied to concrete failure cases.
• Browser testing: provide a test environment and user journey; inspect observed behavior and screenshots.
• Refactoring and migrations: name the pattern to replace and behavior to preserve; review each migration step and its checks.
• Recurring engineering work: define a schedule or repository event, instructions, and limits; inspect each run’s output.

When is Lispa a useful addition?

Consider Lispa when you want a separate repository task to run in the cloud, want several model choices in the same workspace, or want teammates to share an organization credit balance without seat fees. You can use models through Lispa credits; a separate model subscription is optional.

If you already use Claude Code or Codex, Lispa adds a cloud workspace and model routing. It does not lower the fixed price of an existing provider subscription. Lower-cost models may reduce metered model spend, but total cost also includes Lispa’s machine and service charges.

You do not need to know Lisp. Describe the work in plain language. A local Lispa agent is also available for macOS and Linux if you prefer your terminal.

Create an account, choose your plan, then connect a repository. More practical answers.