Skip to content

Releases: SamPlvs/zero-operators

v1.0.2 — First public release

30 Apr 17:22

Choose a tag to compare

v1.0.2 — First public release

Zero Operators (ZO) is an autonomous AI research and engineering team. You write a plan; a team of AI agents builds, trains, validates, and delivers. The delivery repo stays clean — zero ZO artifacts leak in.

This is the first public release.

Highlights

--low-token cost-saving preset

For Anthropic Pro, student, and budget-watching plans. Sonnet lead, Haiku for code-reviewer / test-engineer / oracle-qa, Phase-4 capped at 2 iterations, full-auto gates, earlier auto-compaction. ~30% measured savings on the canonical reference run. CLI flag and plan YAML both supported; override individual knobs with --lead-model, --max-iterations, --no-headlines. (#54#61)

Autonomous experiment loop

Phase 4 auto-iterates in non-supervised modes with target / budget / plateau / dead-end detection. Per-iteration hypothesis → config → metrics → result lineage captured in .zo/experiments/. Plan-level ## Experiment Loop block for project-specific overrides. (#49, #50)

Hard gate enforcement

ZOTrainingCallback is now a contractual requirement — Phase 4 fails if metrics.jsonl and training_status.json aren't produced. Caught the "agent ignored aspirational contract" failure mode. (#59)

Brand v2 + website v2

New palette (canvas / paper / coral / dusk / moss), typography (Geist + Cormorant Garamond + JetBrains Mono), simplified C mark. Astro 5 single-page static, mobile-tested, light/dark theme toggle. (#51)

Platform-aware scaffolding

Docker compose template auto-detects GPU vs CPU host and emits the right runtime. Phase completion snapshots at every gate transition. (#48)

Other fixes

  • MDX 2 parser bugs on the docs site (#62)
  • zo continue --repo, gates set, watch-training correctly thread the delivery-repo hint (#46)
  • macOS Docker compose emits CPU runtime instead of failing on device driver selection (#48)

Full details in CHANGELOG.md.

A note on cost

ZO v1 is optimised for production-grade R&D — agents iterate, verify, and write production-quality code, which burns tokens. A canonical reference run with the default profile is on the order of $10–15 of API credits.

For cost-sensitive plans, use --low-token (~30% measured savings). Path to 70–80% savings via SDK refactor is on the roadmap.

Roadmap

  1. Broader coding-agent provider support — Codex CLI, OpenCode, others as adapters
  2. No-install web UI for non-engineering researchers (biology, finance, social sciences)
  3. Push --low-token toward 70–80% savings via SDK refactor (prompt caching, Batch API, Files API)

Full roadmap: https://docs.zerooperators.com/roadmap

Get started

git clone https://github.com/SamPlvs/zero-operators.git
cd zero-operators
./setup.sh
uv sync --extra dev
zo init my-project
zo draft -p my-project
zo build plans/my-project.md

Documentation: https://docs.zerooperators.com
Issues: https://github.com/SamPlvs/zero-operators/issues

Stats

  • 735 tests passing, 7 skipped
  • 20 agents (11 project delivery + 6 platform build + 3 zo-draft)
  • 24 slash commands across 8 categories
  • 35 PRIORS captured via the self-evolution protocol
  • ruff clean on src/zo/
  • All 8 PRD §9 acceptance criteria met