Injaz Cowork / injazsmart.ai
Injaz Cowork chat
👋 أهلاً! اكتب أي سؤال أو طلب وأنا أرد عليك.
اشرحلي الفرق بين Promise و async/await
اعمل review سريع للكود ده: const sum = arr => arr.reduce((a,b)=>a+b, 0)
how does Injaz Cowork's parallel DAG executor work?

Model-first agent harness, stronger by design.

A CLI agent that classifies, plans, executes, and verifies — using 8 specialized model roles instead of one big agentic loop. Domain-aware verification, parallel DAG execution, and user-approved self-healing.

v0.1 — MVP complete Devops + Browser domains Multi-model routing Parallel + Cross-domain

Why it's different

🎩

8 specialized roles

Understand · Clarify · Plan (3 modes) · Pre-check · Execute · Domain-verify · Final-verify · Memory.

Parallel DAG execution

Independent steps run concurrently in waves. 3-5× faster than sequential agents.

🛡️

Domain-specific verification

Each domain ships its own runtime + final verifier prompts. Per-step semantic checks.

⚕️

User-approved healing

On failure, candidate fixes are gathered from memory + skills. Apply, edit, or halt.

💰

Multi-model routing

Light hats use Haiku, heavy ones use Opus. ~30-40% the cost of Opus-everywhere.

📁

Zero-code domains

Adding a new domain = 4 markdown files. No TypeScript changes required.

Quickstart

# 1. install
git clone <repo> injazcowork && cd injazcowork
npm install && npm run build
export ANTHROPIC_API_KEY=...

# 2. small task — direct exec
node bin/injaz "check disk usage on this server"

# 3. medium task — auto-planned
node bin/injaz "set up nginx with a default config and start it"

# 4. cross-domain plan with DAG visualization
node bin/injaz --mermaid "deploy app to localhost:3000, then go to it in a browser"

# 5. high-assurance prod operation
node bin/injaz --assure full --yes "deploy v2 to production"