Code Beta

A coding agent, with the brakes you choose.

Switch to Code and the agent reads, edits and tests your repository: on your machine through the Windows app, or in your GitHub and Azure DevOps repositories. Every edit and command runs under the permissions you set, and is recorded.

Get the Windows app Start free

renly WorkCode fix/invoice-rounding

Invoice totals are off by a cent sometimes. Find out why and fix it.

Searched the billing module14 files

Reproduced it with a failing test

Rounding happened before tax was applied. I moved it after the tax calculation and added a test for the 0.005 case.

src/billing/totals.ts+12−5

src/billing/tax.ts+3−1

test/billing/totals.test.ts+28−0

$ npm test -- billing
✓ totals rounds after tax (4 ms)
✓ totals handles 0.005 (2 ms)
Tests: 18 passed, 18 total

Approve 3 filesRevert to checkpoint

Illustration. Names and values are representative.

Permission modes

Ask every time, Auto-accept edits, Auto-run allowlisted, Full autonomy or Bypass approvals, chosen per session. Chained commands still ask.

Change sets

Review the proposed edits file by file with line counts, then approve the set in one go.

Checkpoints

A checkpoint before edits means any step can be reverted, so trying an idea costs nothing.

A governed terminal

Build, test and run scripts on your machine. Commands you allow once can be added to an always-allow list for the session.

Sub-agents

Hand investigations to read-only sub-agents that work in parallel and report back, while the main session stays focused.

Git and pull requests

Branches, commits and pull requests in GitHub and Azure DevOps, under the same approvals as everything else.

Bring your setup

Your instructions and tools come with you.

The Windows app imports your Claude Code and Codex CLI instructions and MCP servers, and reads project instructions from the repository, so the agent follows the conventions you already wrote down.

Instructions files such as CLAUDE.md and AGENTS.md become a personal agent you can reuse.
Your MCP servers are imported with the rest of your setup.
Your models: the ones your organisation enables, with GitHub Copilot available as a model provider.

Permissions for this session

Ask every timeEvery edit and command waits for you

Auto-accept editsEdits apply; commands still ask

Auto-run allowlistedCommands on your list run without asking

Full autonomyFor trusted, isolated work

Illustration of the permission control.

Questions, answered.

Where does the code run?

On your own machine through the Renly Windows app, in a folder you choose, or in your GitHub and Azure DevOps repositories through their connectors. Locally, your files stay on your machine and the agent edits them in place.

What can the agent do without asking?

You pick a permission mode for each session: Ask every time, Auto-accept edits, Auto-run allowlisted, Full autonomy, or Bypass approvals. Commands you have not allowed still ask in the stricter modes, and every edit and command is recorded.

Can I undo what the agent changed?

Yes. Renly takes a checkpoint before edits, so you can revert a step, and you review each change set file by file before approving it.

Which models does it use?

The models your organisation has enabled, from Anthropic, OpenAI and Google among others, and GitHub Copilot can be used as a model provider.

Code with an agent your security team can sign off.

Install the Windows app, open a folder, and switch to Code.

Get the Windows app Book a demo