Codoro

Codoro, made by Osokoro, is an agentic IDE for macOS currently in public beta (version 0.2.5). It brings an editor, a design canvas, a project board and team chat into one desktop app, with AI usage governed by spending caps.

What It Is

Codoro is a desktop coding environment where an agent plans multi-file changes, stages them as diffs, and runs allowlisted build or test checks. Nothing is written until the user approves the plan, and optimistic undo restores the reviewed source files unless newer work has changed them. Codoro searches the repository locally and sends only the ranked excerpts a request needs to the selected model.

Four Product Surfaces

• Editor: multi-file edits with approved diffs and undo.
• Design: manifest-declared frame previews bound to repository paths, with tokens read from the stylesheet. Edits arrive as diffs.
• Plan: a milestone board read from git history, showing owners and blockers.
• Chat: repository-scoped channels where the team and the agent share threads.

Models, Caps and Cost Control

Users can switch models mid-task with context carried over, choosing from featured models or any model available through Vercel AI Gateway. Every run reserves its estimated cost before the call and settles against real usage afterwards. A per-run and a per-day cap are required before the first agent run. When a cap is reached, the run pauses and offers a cheaper model instead of downgrading silently. Provider API keys are held in the macOS keychain, or users can sign in with an Osokoro account for hosted credits. Routing rules live in a .codoro/routing.toml file committed with the repository.

Voice Input

After a provider disclosure, the desktop records one spoken turn and transcribes it through OpenAI Whisper. The transcript fills the Agent composer for review and is never auto-submitted. Local macOS speech can read a reviewed diff aloud.

Current Status

The docs describe the current signed public beta for macOS, with downloads per Mac architecture. The homepage's product demo is described as an interactive mock, not a downloadable build.