Cloud Agent Capabilities | Creor

Cloud agents share the same AI engine as the local Creor agent, giving them access to a powerful set of tools for reading, writing, searching, and executing code. This page covers what they can do and where their boundaries are.

Code Generation

Cloud agents can generate new code from scratch based on your prompt. They read the existing codebase to understand patterns, conventions, and dependencies, then produce code that fits naturally into the project.

New features: describe what you want and the agent writes the implementation across multiple files.
Boilerplate: generate API routes, database models, form components, or configuration files.
Migrations: create database migration scripts based on schema changes.
Integrations: add third-party service integrations with proper error handling.

# Example: generate a new API endpoint

curl -X POST https://api.creor.ai/v1/agents/launch \

-H "Authorization: Bearer $CREOR_API_KEY" \

-H "Content-Type: application/json" \

-d '{

"repository": "github.com/acme/backend",

"branch": "main",

"prompt": "Add a POST /api/v1/invitations endpoint that creates a team invitation, sends an email via the existing EmailService, and returns the invitation object. Follow the patterns in the existing /api/v1/users endpoints."

Code Review & Analysis

Cloud agents can review code changes and provide detailed feedback. They analyze diffs against the full codebase context, not just the changed lines.

Bug detection: identify logic errors, off-by-one mistakes, null pointer risks, and race conditions.
Security review: flag potential vulnerabilities like SQL injection, XSS, insecure deserialization, and hardcoded secrets.
Style consistency: check that new code follows project conventions, naming patterns, and architectural boundaries.
Performance: identify N+1 queries, unnecessary allocations, missing indexes, and expensive re-renders.
Type safety: catch type assertion misuse, implicit any usage, and missing null checks in TypeScript codebases.

Tip

For the best review quality, provide context about what the change is intended to do. A prompt like "Review this PR that adds rate limiting to the API gateway" gives the agent domain context that improves its feedback.

Refactoring

Cloud agents can perform large-scale refactoring across your codebase. Because they operate on a cloned copy, there is no risk to your working directory.

Rename across files: rename a function, class, or variable across every file that references it.
Extract patterns: pull repeated code into shared utilities or base classes.
Migrate APIs: update code to use a new API version, library, or framework pattern.
Architecture changes: move files between directories, split large modules, or merge related ones.
Type system upgrades: add TypeScript types to a JavaScript project, tighten loose types, or migrate to stricter tsconfig settings.

Refactoring results are returned as a diff that you can review before applying. The agent does not push changes directly to your repository.

Testing

Cloud agents can generate and run tests for your codebase.

Unit tests: generate test files for existing functions and classes, matching your project's testing framework.
Integration tests: create tests that exercise multiple components together.
Test coverage gaps: analyze existing tests and generate tests for uncovered code paths.
Test fixes: update failing tests after a refactoring or API change.
Test execution: run the test suite inside the sandboxed container and report results.

# Example: generate tests for uncovered code

{

"repository": "github.com/acme/frontend",

"branch": "main",

"prompt": "Find functions in src/utils/ that have no corresponding test file in src/utils/__tests__/. Generate unit tests for them using Vitest, following the patterns in the existing test files."

}

Documentation

Cloud agents can generate and update documentation based on your code.

API documentation: generate OpenAPI specs, endpoint descriptions, and example requests from your route handlers.
Code documentation: add JSDoc, docstrings, or inline comments to public APIs.
README updates: regenerate project READMEs based on current code structure.
Architecture docs: produce Mermaid diagrams or written descriptions of system architecture.
Changelog: summarize recent commits into a human-readable changelog entry.

Available Tools

Cloud agents have access to a subset of the local agent's tools, adapted for the sandboxed cloud environment.

Tool	Available	Notes
read	Yes	Read files from the cloned repository.
write	Yes	Write new files or overwrite existing ones.
edit	Yes	Exact string replacement edits.
multiedit	Yes	Multiple edits in a single tool call.
glob	Yes	File pattern matching.
grep	Yes	Content search with regex support.
ls	Yes	List directory contents.
bash	Limited	Shell commands in a sandboxed environment. No network access by default.
codesearch	Yes	Semantic codebase search (requires RAG plugin).
websearch	No	Disabled in cloud for security.
webfetch	No	Disabled in cloud for security.
task	Yes	Spawn sub-agents for parallel work.
plan	Yes	Create structured plans.
git	Read-only	Git log, diff, and blame. No push.

Note

The bash tool runs in a restricted shell with no outbound network access. Package installation commands that require downloading from registries will fail unless the dependencies are already in the cloned repository (e.g., a committed lock file with a cached node_modules).

Limitations

Cloud agents have some limitations compared to local agents.

Limitation	Reason	Workaround
No internet access from bash	Network isolation for security	Pre-install dependencies before the agent run, or use a Docker image with deps.
No PTY/interactive terminal	Sandboxed environment	Use non-interactive commands only. Avoid tools that require user input.
No persistent storage	Containers are ephemeral	Export artifacts via the API or configure a webhook to receive results.
No IDE integration	Runs headless on server	View results in the dashboard, PR comments, or API responses.
10-minute default timeout	Resource management	Increase timeout in agent settings (max 60 minutes on Pro plan).
Repository size limit	Container disk space	Max 5 GB repository size. Use sparse checkout for monorepos.

Capabilities