For Developers

One command.
Multiple AI agents.

Describe an app — or paste a detailed spec — and multiple AI agents build it end-to-end. Planning, task delegation, implementation, QA loops, and review happen across coordinated stages. One query to a working project.

How It Works

01

Describe what you want

Write a natural-language query or feed in a full spec. 0spec generates a multi-agent flow that can build the whole project end-to-end: architecture, delegated implementation tasks, QA loops, and review. Attach mockups or docs with --ref for richer context. Use --chat to refine the plan before agents start working.

02

Agents plan and execute

Multiple AI agents work in parallel: one plans the architecture, another writes code, a third reviews for bugs. Each agent has the right tools for its job.

03

Review the results

Every session produces structured artifacts: plans, code, reviews. Resume any session, inspect stage outputs, or re-run from a specific point.

What You Can Do

Build a full feature

Describe the feature. Agents plan the architecture, implement it across multiple files, write tests, and produce a review. One command, start to finish.

0spec --query "add OAuth login with Google and GitHub"

Fix a complex bug

Point agents at the bug. They read the codebase, trace the issue, implement the fix, verify it doesn't regress, and explain what went wrong.

0spec --query "fix race condition in session manager" --flow fix-review

Refactor safely

Agents plan the refactor, implement it incrementally, run tests at each step, and produce a diff review. Git worktree isolation means your working tree stays clean.

0spec --query "extract auth middleware into shared package"

Code review

Run multiple review agents in parallel: one checks correctness, another checks security, a third checks test coverage. Each agent has its own skillset — custom tools and instructions tailored to its role.

0spec --task "review last 5 commits" --run-agent reviewer-claude

PR review on every push

Trigger 0spec from a GitHub webhook, MCP, or your CI pipeline. Agents review the diff, post inline comments, and flag regressions — automatically on every PR.

0spec --task "review PR #42" --run-agent reviewer-claude --run-agent security-claude

Run anything on a cron

Schedule any flow to run on a recurring basis. Nightly QA sweeps, weekly dependency audits, daily competitive research with open-source models — set it and forget it. Available on Solo and Pro plans.

0spec --query "audit deps for CVEs" --cron "0 3 * * *"

Why Multiple Agents

Specialization

A planning agent thinks differently than an implementation agent. Separate roles produce better output than one agent doing everything.

Cross-checking

Parallel review agents catch different classes of bugs. Claude finds logic errors. Codex finds integration issues. Together they cover more ground.

Speed

Independent work runs in parallel. A 10-stage flow with 5 parallelizable stages finishes in the time of 6, not 10.

Questions

How does it go from a description to a working app?

You provide a query or a spec. The planner agent analyzes it and generates a staged flow with architecture, delegated implementation tasks, QA loops, and review. Agents pass artifacts between stages, iterate when checks fail, and keep driving toward a working, reviewed app. Attach mockups or docs with --ref for richer context and use --chat to refine the plan before execution.

What happens if an agent fails mid-run?

The session state is saved. Run --resume with the session ID and 0spec picks up from the failed stage. No wasted work.

Can I use just Claude without Codex?

Yes. You can run just Claude, just Codex, or mix engines. Plan limits control engine count, providers, and concurrency, not a specific engine choice.

Does it modify my files directly?

By default, each session runs in an isolated git worktree. Your working directory is untouched until you merge the results.