Salacia is the operating harness for AI coding agents: it builds the right context, enforces boundaries, verifies the outcome, and only then promotes the patch.
Claude Code · Codex · Cursor · Cline · OpenCode · Antigravity
Goal, mutable surface, verification, promotion policy.
Repo map, working set, history, guardrails.
Accept, reject, or block with traceable evidence.
Real Artifact
Not a hand-drawn mock. A real trace from a blueprint-backed run.
Why Salacia
Salacia turns a raw agent loop into an engineering system: bounded context, explicit policy, hard verification, and a real promotion decision.
Choose Your Lens
This is where Salacia stops looking like a demo and starts looking like infrastructure.
Use Salacia when the model is smart enough to solve the task, but still wastes time loading the wrong files, changing too much, or shipping changes without a clear verdict.
If you are building an agent, IDE bridge, or internal assistant, Salacia gives you a control plane, a context plane, and a judge loop without forcing you to replace your agent model.
Salacia is useful when your problem is no longer “can the model code,” but “how do we make these runs reviewable, comparable, and safe to promote.”
How It Works
One control file. One bounded run. One verdict.
Declare the goal, mutable surface, verification, and promotion policy in a form the runtime can enforce.
npx salacia init
Compile the program into a blueprint and build a bounded ContextPack instead of dumping the whole repo into the prompt.
salacia design
Dispatch the agent in an isolated workspace with the generated context, budget, and guardrails.
salacia run
Accept, reject, or block the patch, then inspect the full trace and evidence behind that decision.
salacia judge && salacia trace
Open source. Apache 2.0. Built for teams that want coding agents to behave like engineering systems, not demos.