zecor.dev
Ship features.
Spend tokens last.
Zecor is an autonomous software construction engine that pairs local deterministic execution with hybrid model routing to build and verify software with minimal token overhead.
Install
git clone https://github.com/zecordev/zecor && cd zecor uv sync --group dev cp zecor.config.example.json zecor.config.json uv run zecor doctor
Requires Python 3.12+, git, ripgrep,
ast-grep, and any OpenAI-compatible local model endpoint.
What it is not
Not a chat wrapper. The exit code of a deterministic gate is the sole definition of done; escalation is driven by observed artifacts — a still-red gate, an unchanged failing-assertion set, a duplicate diff, scope creep — not by a model's self-reported confidence.
Rust internals, published
The deterministic layer leans on small, fast Rust tools. They stand alone and are published under the zecordev org.
Cross-platform process isolation for untrusted test execution: one policy — Landlock/seccomp on Linux, sandbox-exec on macOS, Job Objects on Windows.
zecor-tokcountExact BPE token counting and byte-precise context trimming for LLM cost control.
Also public: zecor-secretscan, zecor-astscan, zecor-gitengine.
Read next
- Architecture — the gate, the routing ladder, the triggers.
- Security — autonomy modes, sandboxing, injection handling.
- Configuration — local runtimes, tool paths, limits.
- Benchmarks — cost per feature, hybrid vs pure cloud.