Arc
+ ++ Open Source +
+
+ The AI software factory for
+ expert engineering teams
+
+ + Ship production code through long-horizon deterministic workflows, built-in verification, + and full observability — without trading quality for speed. +
+ ++ Controlled, auditable pipelines for real engineering +
++ Instead of one-shot code generation, Arc runs workflows: plan, implement, verify, review, ship — using + the right model at each step, with guardrails that fail closed. +
+Deterministic long-horizon runs
++ Predictable automation with guardrails and rerun-ability. Human-in-the-loop when needed, fully automated when it's not. +
+Verification-first pipeline
++ Builds, tests, static analysis, LLM review, and human approval — baked into every workflow, not bolted on after. +
+Multi-model orchestration
++ Use the right model for each step. CSS-like stylesheets assign fast models for breadth, frontier models for correctness. +
+Total observability
++ Trace every step down to each tool call. Inspect inputs, outputs, and timing. Compare runs and diagnose regressions. +
+Cloud sandboxes
++ Ephemeral sandboxes for safe execution. Preview environments to validate changes end-to-end. SSH access when you need it. +
+Open source, forkable
++ Written in Rust. AGPL-licensed. Fork it, extend it, make it the center of your engineering process. +
++ Built for real engineering, not promptware +
++ Arc is built like infrastructure: solid architecture, strong defaults, and the reliability you expect from production tooling. +
+-
+
- + + Written in Rust for performance, safety, and a durable core + +
- + + Designed for long-horizon tasks across real codebases, not toy demos + +
- + + Open source — fork it, extend it, and make it the center of your business + +
Hard gate. Fails block the pipeline.
+Policy checks enforced at every stage.
+Additional layer, not a substitute for real checks.
+Integrated cleanly into the workflow graph.
++ Verification is a first-class concept +
++ Arc doesn't just produce output — it validates it. Builds and tests are gates, not suggestions. + LLM review adds a layer. Human review loops close the gap. +
++ Mergeable results that survive CI and code review. +
++ Multi-model, multi-provider +
++ Different steps deserve different models. Arc orchestrates an ensemble so you get the best + quality per dollar — fast models for breadth, frontier models for correctness. +
+-
+
- + + CSS-like stylesheets assign models to workflow stages + +
- + + Provider flexibility — avoid lock-in, keep leverage + +
- + + Ensemble strategies to maximize quality and minimize cost + +
+ Ground-up observability +
++ If you can't see what happened, you can't trust it. Full traceability across runs — inspect every model + call, every tool invocation, every decision point. Compare runs, diagnose regressions, tune workflows. +
++ Shippable outcomes, not impressive output +
++ Most AI coding tools optimize for impressive output. Arc optimizes for what actually merges. +
++ Workflow-as-Code +
++ Define how work gets done in version-controlled graph files, then run it consistently. + Composable steps. Auditable history. Repeatable results. IaC for coding. +
+