Faelith Code
Ship the change. Keep the judgment.
An agent that reads your whole repository, proposes the diff, runs your tests and explains every hunk before a single line lands. You stay the reviewer; it does the typing.
Try it right here
This is the real Faelith Code app. Type a request, open Changes, Files or the sidebar, switch modes and models.
Why teams hand it the hard tickets
It reads before it writes
The agent walks the tree, follows the call sites and opens the tests first, so the diff fits your codebase instead of a tutorial.
Every hunk comes with a reason
Accept, reject or ask why, one hunk at a time. Nothing is applied until you say so, and nothing is hidden behind "trust me".
Big changes, isolated
Parallel splits a migration across subagents, each in its own git worktree. Peak runs several implementations and keeps the best one.
From ticket to merged, in three moves
- 01
Describe the outcome
Paste the issue or say what should be true when it is done. Plan mode writes the approach for you to approve first.
- 02
Review the diff
Hunks arrive with the reasoning attached. Push back in plain language and the agent revises.
- 03
Let it prove it
It runs your test suite and shows the result. Green means done; red means it keeps going.
Questions people ask before switching
Does it change files without asking?
Only within the permission settings you choose. Plan and Ask modes never edit or run anything.
Which models does it use?
Echo for fast everyday work and Horizon for deeper reasoning, each with a 1M-token context window.
How is it billed?
Plans include a 5-hour and weekly allowance shared with the CLI and Chat. You always see how much is left.
Your next forty-file refactor is an afternoon.
Install the desktop app, open a repository and hand over the ticket you have been avoiding.