Cloud Agents Overview

Cloud agents are AI agents that run on Creor's infrastructure instead of your local machine. They can operate autonomously on your repositories -- reviewing pull requests, detecting bugs, generating documentation, and running long-lived coding tasks without tying up your IDE.

What Are Cloud Agents

A cloud agent is an instance of the Creor AI engine running on Creor's servers. It clones your repository into a sandboxed environment, executes a task using the same tool-use loop as the local agent, and reports results back to you via the dashboard, API, or webhooks.

Cloud agents are stateless per invocation -- each run starts with a fresh clone of your repository at the specified commit or branch. They have read and write access to the cloned workspace, can run shell commands in a sandboxed container, and can interact with the LLM to reason about code.

Note

Cloud agents never push changes directly to your repository. They produce artifacts (diffs, reports, PR comments) that you review and apply at your discretion.

Use Cases

Cloud agents excel at tasks that are triggered by events (a new PR, a scheduled review) or that take a long time to run.

Use CaseTriggerOutput
Automated code reviewNew pull requestInline comments on the PR with suggestions and bug reports
Bug detection (Bugbot)PR opened or updatedComments flagging potential bugs, logic errors, and security issues
Test generationManual or scheduledNew test files committed as a draft PR
Documentation generationManual or scheduledUpdated README, JSDoc, or API docs
RefactoringManualDiff output showing proposed changes across multiple files
CI/CD integrationBuild failureAnalysis of the failure with a proposed fix
Dependency updatesScheduled (weekly)PR updating dependencies with compatibility notes

Cloud vs Local Agents

Both cloud and local agents use the same Creor engine and tool system. The differences are in where they run, what they can access, and how they are triggered.

FeatureLocal AgentCloud Agent
Runs onYour machineCreor servers
TriggerYour chat messagesAPI calls, PR events, schedules, dashboard
Repository accessYour local filesystemCloned copy in sandboxed container
Internet accessFull (your network)Restricted to allowed domains
Tool accessAll 25+ toolsSubset (no PTY, limited bash)
Session persistencePersistent across restartsPer-invocation, stateless
IDE integrationFull (inline edits, diff views)Results via dashboard, API, or webhooks
CostUses your LLM API keys or Creor creditsUses Creor credits (compute + LLM)

Tip

Use local agents for interactive work where you want to iterate quickly. Use cloud agents for automated, event-driven tasks that should run without your involvement.

Architecture

When a cloud agent is launched, the following happens behind the scenes.

1. API request or event trigger received
2. Sandboxed container provisioned with resource limits
3. Repository cloned at specified branch/commit
4. Creor engine started inside the container
5. System prompt assembled with task instructions
6. Agent runs tool-use loop (read, edit, bash, grep, etc.)
7. Artifacts collected (diffs, reports, comments)
8. Results delivered via API response, webhook, or PR comments
9. Container destroyed, no persistent state retained

Each cloud agent run is fully isolated. Containers do not share filesystems, network namespaces, or process spaces. The repository clone is destroyed after the run completes.

Pricing

Cloud agent usage is billed from your Creor credit balance. Costs include both the LLM inference (input/output tokens) and a small compute surcharge for the sandboxed container.

ComponentCostNotes
LLM tokensSame as gateway pricingVaries by model. See Models & Pricing page.
Compute$0.01 per minuteBilled per minute of container runtime, rounded up.
StorageFreeTemporary clone. No persistent storage charges.

A typical code review on a medium-sized PR (500 lines changed) costs approximately $0.05-0.15 depending on the model used. Long-running tasks like full-codebase refactoring will cost more due to extended compute time and higher token usage.

Next Steps