Start with project memory. Add agents, expert workflows, human oversight, and secure runtime. Every part is visible, replaceable, and yours.
AIWG turns hard-won ways of working into inspectable building blocks. Use the complete system or start with one capability. The project—not a vendor account—owns the result.
Keep decisions, architecture, instructions, and operating context coherent across providers.
Turn reliable methods into reusable agents, skills, commands, reviews, and complete workflows.
Make status, decisions, evidence, approvals, and failures visible to the people responsible.
Run long-lived agents through Cockpit and Agentic Sandbox on infrastructure you control.
Open your project in a supported AI provider and paste the instruction below. The installer inspects first, explains its plan, preserves your work, installs AIWG, and verifies the result.
Open your project in Claude Code, Codex, Cursor, Copilot, Warp, or another supported provider.
Give the provider the setup manifest and let it inspect the current system.
AIWG builds context, deploys the toolkit, and proves that the provider is engaged.
Install or repair AIWG for this project by following https://aiwg.io/setup.aiwg.yaml Explain the plan before changing anything, preserve my existing work, and ask me only for choices you cannot safely determine.
AIWG is open source. It maintains one provider-neutral source and deploys agents, skills, rules and MCP tools into the paths each AI tool reads, with reusable project context and SDLC workflows. Switching providers is configuration, not a rewrite.
Project-local extensions, addons, frameworks, and plugins keep the same shape as upstream artifacts. Build locally, share privately, or contribute publicly without translating the work.
Explore open customization →The optional decision-engine addon sends bounded questions to Jev, a decision service: yes/no, a pick from known options, or a 1–5 score.
It returns a typed answer with a confidence score.
If Jev abstains, times out, is not configured, or falls below the confidence threshold, your assistant decides as usual.
aiwg use decision-engine
aiwg decision setup jev --token-stdin --verify
aiwg decision ask --question "..." --yes-no --json
Decision support is opt-in; aiwg use all does not install it. Jev answers are
advisory, never authorization.
Savings vary by workload; it helps most with many small bounded questions. Questions and context are sent to Jev, so keep secrets out of them.
Essays, technical guides, research notes, and release reports from building durable AI-assisted systems in the open.
Why useful practice must survive changes in project, provider, and team—and how identical-form artifacts make that possible.
Read the essay →What people must be able to see, approve, and stop.
Read →Building a coherent system prompt from project-owned sources.
Read →AIWG is MIT licensed because advanced capability should be broadly available. Inspect every part, adapt it for your own work, and share improvements on your terms.
Create project-owned skills, agents, commands, rules, and complete bundles without maintaining a fork.
Join the community, compare methods, report gaps, and help make high-quality AI practice accessible.
Promote useful project-local work into the shared corpus without rewriting its structure.
Add AIWG Cockpit and Agentic Sandbox when work needs isolation, durable sessions, human approval, and an audit trail. Use containers for speed or VMs for a separate kernel.
Copy the secure setup promptObserve sessions, approve decisions, inspect state, and stop work.
Self-hosted container and VM execution with resource policy and audit evidence.