Code with agents.
Keep the final say.
A free, open-source AI coding workspace with visible agent actions, approvals, and reviewable changes.
Product preview — Nexora desktop app
Free · Open source · Local-first · Bring your own models
The agent run
One run, from read to review.
Nexora shows file reads, tool calls, commands, and edits as one continuous run. Dangerous shell commands always stop for your explicit approval — in every autonomy mode — before the agent continues. Every change ends in a reviewable diff tied to the run that produced it.
Illustrative interaction demo — synthetic example-app data, not real Nexora performance
The agent reads the project files before it touches anything.
src/auth.ts · src/auth.test.ts
Bring your own models
Use your models, not ours.
Connect your own provider, or point Nexora at any OpenAI-compatible endpoint. Credentials are stored in the OS keyring — never on a Nexora server, because there isn’t one.
- OpenAI
- Anthropic
- Gemini
- OpenRouter
- NVIDIA NIM
- OpenCode Zen
OpenAI-compatible endpoint · credentials stored in the OS keyring
No black box
No black box between you and your agents.
Inspect the code, bring your own models, own your workflow. Your workspace stays local, agent actions stay visible, and credentials stay in the OS keyring.
MIT licensed · open source on GitHub · no account required
Trust should be inspectable.
Read how Nexora handles data, provider requests, permissions, and known limitations.
Download Nexora
Prebuilt Windows installer, or build Nexora from source. All releases are on GitHub.