Zero Config
Detects your stack automatically. No YAML, no JSON, no setup files. Run init in any project — new or existing — and the environment is ready.
Always Improving Factory
ai transforms into always improving
Zero-config AI development environment. Detects your stack, installs skills, configures agents — then drives a spec-driven loop that plans, implements, verifies and evolves.
scroll to discover the workflow
The development loop
01
Planning, simplified
/aif-plan · /aif-improve
One command triggers planning. A milestone becomes a branch, detailed tasks and dependencies — refined for gaps before a single line is written.
02
Execution with checkpoints
/aif-implement · /aif-fix
Tasks execute one by one with commit checkpoints. When something breaks, the fix command reads the plan and knows exactly what was intended.
03
Verified before merge
/aif-verify · /aif-qa · /aif-loop
Every task is validated against the plan, then a three-stage QA pipeline produces a risk-annotated change summary, test plan and concrete test cases. The reflex loop refines until it passes.
04
Always improving
/aif-evolve
Accumulated patches are analyzed and folded back into skill capabilities. The factory gets measurably better with every cycle you run.
Designed for today's development,
beyond copy-paste prompting.
Spec-Driven Development
AI follows a plan, not random exploration. Persistent specs — roadmap, architecture, rules — make every workflow predictable, resumable and reviewable.
One Factory, 15+ Agents
Claude Code, Cursor, Windsurf, Copilot, Codex, Gemini CLI, Antigravity, Roo Code, Warp, Junie and more. One workflow, any agent your team already uses.
Security First
Two-level security scanning for every external skill from the skills.sh marketplace. Community-built extensions, safe by default.
QA Pipelinenew
/aif-qa runs a three-stage manual-testing workflow: risk-annotated change summary, scoped test plan, and concrete TC-NNN test cases — artifacts saved per branch.
Skill Distillationnew
/aif-distillation turns books, docs, folders or URLs into reusable Agent Skills — workflows, heuristics, checklists and examples instead of a long summary. Split mode builds a whole toolkit.
More tokens now.
Less pain later.
Structured workflow costs ~20% more upfront on planning — and saves ~60% on fixes, rewrites and context recovery.
Without workflow
- Token usage“Cheap” start, expensive rewrites
- After featureManually update docs, CI, contracts, configs
- ContextAgent forgets decisions between sessions
- Bug fixingRe-explain everything from scratch each time
- QualityRandom results, no verification step
With AI Factory
- Token usageInvest in planning, save on rewrites and fixes
- After featureDocs, CI, contracts — updated inside the cycle
- ContextPersistent specs: roadmap, plan, architecture
- Bug fixing/aif-fix reads the plan and knows the intent
- Quality/aif-verify checks every task before commit
From zero to production.
The complete path for any new project. Clear context between steps — each step generates everything the next one needs.
Prepare the project
Run this once in a repository to install AI Factory and create the shared project context.
- npm i -g ai-factoryInstall globally
- ai-factory initDetect stack, install skills, configure MCP servers
- /aifInitialize project structure and primary configuration
Shape the work
Turn requirements into clear milestones before implementation starts.
- requirementsThe most important step. Quality of output = quality of input
- /aif-roadmapDecompose requirements into milestones with clear deliverables
Build each milestone
Repeat this loop for every milestone until the work is verified and ready to merge.
- /aif-planBranch, detailed tasks, dependencies
- /aif-improveRefine the plan — fix gaps, verify dependencies
- /aif-implementExecute task by task with commit checkpoints
- /aif-verifyValidate: everything done, nothing forgotten
- /aif-qa --allChange summary → test plan → test cases for manual QA
- /aif-fix · /aif-commitProblems? Fix them. All good? Commit and merge
Autonomous kanban. Zero oversight.
Create a task — AI plans, implements and reviews it. A fully autonomous pipeline built on AI Factory and the Claude Agent SDK: agent orchestration, self-healing loops, layer-aware parallel execution, real-time board.
Live task pipeline
Plan ready
Scope, dependencies and acceptance criteria are prepared.
Implementing
Agent executes tasks with checkpoints and recovery loops.
Review
Changes are verified before the task moves to done.
How the factory works and how it ships production-grade code.
Planning takes ~20% more tokens upfront. But without a plan, agents waste tokens on wrong approaches, context recovery and rewrites. The structured workflow pays for itself within the first 2–3 features.
Yes. Run ai-factory init in any project — it detects your stack, installs relevant skills and configures MCP servers. Existing code stays untouched. You get the workflow on top of what you already have.
/aif-verify catches missed tasks and deviations from the plan. If something is wrong — /aif-fix reads the plan and knows exactly what was intended. Each feature runs on a separate branch, so main stays clean.
No. The workflow is modular — skip what you don't need. Small fix? Just /aif-fix and /aif-commit. Big feature? Use the full cycle. The only step that's always recommended is /aif-verify before commit.
Initialize AI Factory separately in each service — every microservice gets its own init with its own description, architecture and skills. Then add shared rules via /aif-rules describing how services communicate: API contracts, event schemas, shared types.
Use /aif-rules and add a rule like “after implementation, update API_REFERENCE.md and CHANGELOG.md”. Rules run automatically after each implementation cycle — your custom docs never drift.
Stop configuring. Start building.
Join 1000+ developers shipping with AI Factory.
1000+GitHub stars
15+AI agents
20+skills
MITlicense