v0.1.1 · one native Rust binary
The terminal coding agent that proves its work.
Ask most agents to fix a bug and you get “Done!”. Ask zerd and you get what actually happened: the checks that ran, a second model’s review of the real diff, and every file path looked up on disk.
curl -fsSL https://zerd.si/install.sh | sh
Windows: irm https://zerd.si/install.ps1 | iex
A real session, not a mock-up.
Checks, then a second opinion
Runs the checks that cover your changed files, then a different model reviews the real diff.
Findings get cross-examined
A verifier tries to refute each review finding. High-severity claims are reproduced in a sandbox or capped.
No invented paths
Every path:line in the answer is looked up in your tree before you see it.
Cheap by design
Remembers what it read and sends only new lines or a diff. 85% of input comes from cache in a real session.
Rules enforced in code
Confined tools, a path-scanning shell and an OS sandbox, not a prompt asking nicely.
Fails over, and says so
When a provider hits its quota zerd switches to another one and tells you.
Works with the models you already use
Claude, GPT, Codex, DeepSeek and local models. macOS, Linux and Windows.
Models and routing