ArxCode Studio
The flagship — a sealed agentic coding system that plans, builds, verifies, and ships working code.
Studio is the flagship builder. It is an agentic coding system, not a model wrapper: capability comes from the harness — an engineered scaffold of tools, a controlled execution loop, and hard verification gates — that drives a model to ship. The model proposes; the harness grounds, runs, checks, and corrects. The whole loop runs inside the sealed boundary, so your code is never retained and never trains a model.
How a build works
A build is not a single prompt and response. It is an agent loop the harness runs until the work is verifiably done:
- Plan. The agent decomposes the goal into steps and produces an impact summary before touching anything.
- Act. It calls typed tools — read file, write file, run command, search repo, run tests — through a structured interface.
- Observe. Tool results (compiler errors, test output, diffs) feed back as grounded observations, not guesses.
- Verify. Gates run automatically: build passes, tests green, no new lint or type errors. Failures loop back to Act.
- Settle. Only a verified result is returned, and the consumed
$ARXis burned.
A raw model that "writes code" hallucinates APIs, skips edge cases, and can't tell whether its output runs. The harness closes that gap with real tools, real feedback, and hard verification gates — so "zero new bugs" is enforced by the loop, not promised.
Features
- Typed tool layer — read, write, exec, search, test — sandboxed inside the boundary.
- Verification gates — build, test, and lint must pass before a result is accepted.
- Self-correction — a failed gate feeds the error back as an observation and the agent retries.
- Context engineering — repo-aware retrieval and compaction keep the agent grounded across long builds.
- Multi-file, full-stack edits with dependency-aware impact analysis up front.
- Live preview for UI builds, served from inside the boundary.
- Language-agnostic — TypeScript, Go, Python, Rust, and more.
- Model-portable — the harness runs against swappable open models with no product change.
- Per-build
$ARXmetering with the cost shown up front.
Chains into
A Studio build flows naturally into ArxTest for coverage and ArxAudit for a security pass — all in the same sealed session.
Next
- Walk through a real build in Run your first build.