Skip to main content
Voss is an agent engineering organization layer: it runs AI coding agents like an engineering team — declared roles, hard budgets, scoped tools, independent review, and a replayable trail of every action. It has two surfaces. The harness is the bounded agent loop for a single repo task — plan, execute, inspect, resume. The orchestration layer runs a whole team from one goal, with a board, a session tree, and independent reviewers gating completion. Both are declared and controlled through the .voss language.

Run a team

Use voss team run to execute a goal as a reviewed team run.

Run your first task

Use voss do for a bounded one-shot repo task.

Make a scoped edit

Use voss edit to constrain writes to a target and related tests.

Learn .voss

See how the language expresses control — for one agent or a team.

​
Run a team

A team is declared once in .voss, then a single goal runs as scoped, budgeted, independently reviewed work:

​
What Voss is for

Voss is for developers who already use AI coding tools and want stronger boundaries:
  • Clear separation between planning and mutation
  • Permission prompts for risky operations
  • Scoped file writes for edit sessions
  • Resumable project context
  • Code-aware recall across rebuildable code cache and curated project memory
  • Conservative context packing with an explicit --no-pack escape hatch
  • Workflow rules that can move from prompts into .voss code
  • Multi-agent runs that stay scoped, budgeted, and independently reviewed
Voss v0.1 is harness-led and ships through npm as @vosslang/cli. Rust, Homebrew, editor marketplace work, hosted teams, and broad launch polish remain deferred.

​
What to read first

  1. Install Voss: Install Voss
  2. Run a read-only one-shot task: First task
  3. Try a scoped edit session: First edit
  4. Learn the safety model: Modes and Security
  5. Run a whole team: Orchestration overview