AgentJr — The AI Junior Developer That Manages Your Entire Freelance Business While You Sleep
Most "AI developer tools" do one thing: write code.
AgentJr does everything a real junior developer does.
Code. Clients. Communication. Deployment. Testing. Invoices. Social media. All of it. Automatically. While you sleep.
The Real Problem With AI Coding Tools Today
Claude Code writes code. Devin writes code. Copilot writes code.
But writing code is maybe 40% of what a freelance developer actually does.
The other 60%?
- Talking to clients. Understanding what they actually want.
- Managing git properly. Branches, commits, PRs.
- Running tests. Catching bugs before the client sees them.
- Deploying to the right environment. Not accidentally pushing dev code to production.
- Sending updates. "Hey, your feature is live."
- Tracking costs. How much did this project actually cost me in API calls?
- Following up. Drafting invoices. Scheduling calls.
- Posting on LinkedIn about the work you just shipped.
No AI tool today handles all of that together. They give you a coding assistant and leave the rest to you.
AgentJr is different. AgentJr is not a coding assistant. It's a complete AI junior developer that manages your entire workflow — from the moment a client messages you, to the moment the project is deployed, tested, and the invoice is sent.
You Are the CEO. AgentJr Is the Manager.
Here's the architecture that makes AgentJr unique:
You — give direction. Set priorities. Approve plans. That's it.
AgentJr (Manager) — understands requirements, asks smart questions, builds plans, manages git, monitors work, handles client communication, runs tests, manages deployments, tracks costs, drafts invoices, posts social media updates.
Claude Code / Codex / Gemini CLI (Worker) — writes the actual code. Spawned by AgentJr on your terminal. Per project. You choose which one.
AgentJr never writes code itself. It orchestrates the worker that does. This separation is intentional — and it's what makes the whole system reliable.
How It Actually Works — The 21-Step Flow
Every project — from every client, from every channel — goes through the same complete flow:
Step 1 — Message arrives
A client messages you on WhatsApp, Telegram, Email, Slack, or the dashboard. AgentJr receives it immediately.
Step 2 — Content moderation
Is this a normal request? A frustrated client? An abusive message? AgentJr handles each differently — professionally, automatically.
Step 3 — Client check
Known client? Match to their existing project. New client? Ask you for approval first. Unknown contact — nothing happens without your sign-off.
Step 4 — Infinite Clarity Loop
AgentJr asks smart, context-aware questions — one by one — until it reaches 100% confidence. No assumptions. No hallucinated features. It asks about requirements, tech stack, edge cases, timeline, everything. The loop has no limit. It stops only when it's truly ready.
Step 5 — Cost + Timeline estimate
Before any plan, AgentJr auto-generates a cost and timeline estimate. You approve it. The client doesn't see it until you say so.
Step 6 — Client onboarding
New client? AgentJr welcomes them, collects their profile, sets their preferred notification channel. First impression handled professionally.
Step 7 — Plan Mode
AgentJr proposes a complete task breakdown with timeline estimates for each task. You review it. Modify it. Approve it. Not a single line of code gets written until you approve the plan. This is collaborative AI — not AI going rogue on your codebase.
Steps 8–10 — Project setup
Folder structure created. AGENT.md, MEMORY.md, SPEC.md generated. Git repo created or connected. Worktrees set up if parallel tasks needed. The correct vault (dev/staging/prod) injected automatically.
Step 11 — Spawn the coding worker
Claude Code, Codex, or Gemini CLI — whichever you configured for this project — spawns on your terminal. It works task by task, following the approved plan.
Steps 12–13 — Monitor + Auto-resume
AgentJr pipes stdout and stderr in real time. Pattern detection running. Watchdog active — if the coding worker goes silent for 60 minutes, AgentJr detects it, kills it, restarts it, resumes the same session. You don't have to watch anything.
Step 14 — Error handling
Error occurs? AgentJr tries to self-fix. If it can't, it retries via the coding worker. 10 retry limit. $5 cost guard. If still failing, it escalates to you — not before.
Step 15 — Security scan
After every single task — automatically — AgentJr scans the generated code. JWT checks. Rate limits. Access control. Every time. Not just at the end.
Steps 16–17 — Tests + Commit
Playwright MCP server runs browser tests on the live URL. Unlimited retry with cost guard. Tests pass? AgentJr commits to a branch — never directly to main. Tests fail? It goes back to fix, not forward to commit.
Steps 18–19 — Deploy + Health check
AgentJr asks you which environment. You say staging or prod. It deploys. Immediately runs a Playwright health check on the live URL. Health check fails? Auto rollback. You never have to manually rollback.
Steps 20–21 — Notify + Log
You get notified. Client gets notified on their preferred channel. Every task logged with screenshots, asciinema recordings, Playwright videos, and full cost tracking.
AgentJr Talks to Your Clients For You
This is where AgentJr goes far beyond any coding tool.
Your clients can message AgentJr directly — on Telegram, WhatsApp, Gmail, Slack, or Discord — and AgentJr responds as your junior developer.
Client on WhatsApp: "Hey, can you add a dark mode to the app?"
AgentJr: "Got it! Let me ask a few quick questions before I start..."
AgentJr introduces itself as a junior developer from your team. It's professional. It's patient. It handles the whole conversation.
It detects what language the client writes in — and replies in the same language automatically.
It knows when to be quiet. While work is happening, clients don't get spammed with updates. They get one message when the work is done and live.
It handles difficult clients too. Content moderation built in — abusive messages get flagged, you get alerted, and you decide whether to block.
Per-client settings: notification preferences, quiet hours, rate limiting, blacklist option. Every client handled exactly how you want.
Multiple Clients. Multiple Projects. Simultaneously.
Three clients message at the same time. Here's what happens:
Queue (BullMQ + Redis):
1. You (Browser, HIGH priority) → new project for Ali
2. Ahmed (WhatsApp, NORMAL) → add dark mode
3. Sarah (Telegram, NORMAL) → fix login bug
AgentJr processes each in ~200ms.
Meanwhile, in the background:
- Terminal 1: Claude Code building Ahmed's dark mode
- Terminal 2: Claude Code fixing Sarah's login bug
- Terminal 3: Ali's project in Clarity Loop
Nothing missed. Nothing blocked. Nothing confused.
Git worktrees handle parallel tasks on the same project too. Two features needed at once? Two separate worktrees. Two separate Claude Code instances. Both running. Both tracked. Both tested independently. Merged to main only after both pass.
The Memory That Never Forgets
AgentJr has three levels of memory — and they all stay forever:
AGENT.md — You write this. Your rules, your coding standards, your preferences, your client communication style. AgentJr follows them on every project.
MEMORY.md — AgentJr writes this automatically. Every project teaches AgentJr something. What worked. What didn't. What this client prefers. What this tech stack needs. It gets smarter with every project it completes.
Vector Search (FAISS) — Semantic search across all past context. "How did we handle authentication for that last project?" AgentJr finds it in seconds.
Three levels of hierarchy: Global → Project → Task. The right rules apply at the right scope. Always.
Your Secrets Are Safe
AgentJr has a 3-layer vault system. AES-256 encryption. .env files never permanently written to disk — injected at runtime, deleted after the task.
Layer 1: AgentJr's own keys. Your Claude API key, GitHub token, Telegram bot. Never accessible to projects.
Layer 2: Per-project, per-environment vaults. vault.dev.enc uses mock AI and fake payments. vault.staging.enc uses cheap models and test Stripe keys. vault.prod.enc uses your best model and real keys — with extra confirmation required before touching.
Layer 3: Mock mode. Testing costs zero. No real API calls. No real money.
Every inbound webhook is HMAC verified. GitHub, Stripe, custom — all verified before processing. Unverified request? 401 + logged + you get alerted.
Business Features Built In
AgentJr isn't just for code. It handles the business side too.
Social media — AgentJr drafts posts for LinkedIn, Twitter/X, Instagram, Facebook. Sends you the draft for approval. You approve, reject with feedback, or let it auto-post after 3 days. Posts go to all platforms simultaneously.
Invoicing — AgentJr drafts the invoice when a project is done. Stripe + PayPal supported. It never sends without your explicit CONFIRM. Always your call.
Lead generation — Apollo.io API integration. Legal. Targeted. You set the criteria.
Calendar — Google Calendar check before any scheduling. No blind booking.
Estimates — Cost and timeline estimate generated before any project starts. Yours to approve before the client ever sees it.
What's Under the Hood
For the technically curious:
| Layer | Technology |
|---|---|
| AI Brain | LiteLLM — Claude, GPT-4o, Gemini, Groq, DeepSeek, Ollama, any model |
| Queue | BullMQ + Redis — 4 priority levels |
| Terminal UI | Ink.js — rich interactive CLI |
| Dashboard | Next.js + Socket.io — real-time web UI |
| Mobile | React Native + Expo — iOS + Android |
| Database | SQLite (local) + PostgreSQL (cloud) |
| Vector Memory | FAISS / Chroma |
| Testing | Playwright MCP Server |
| Encryption | AES-256 |
| Process Manager | PM2 — 24/7 auto-restart |
| Observability | Prometheus metrics + optional Grafana |
| Extensions | MCP (13,000+ servers) + Skills + Hooks + Plugins |
Model-agnostic by design. You bring your own API key. You choose the model per project. Complex project? Use Claude Opus. Simple task? Use Groq — fast and cheap. Offline? Ollama runs locally, free, private.
Smart fallback chain: primary model fails → fallback 1 → fallback 2 → local Ollama → notify you. AgentJr never stops because one provider is down.
One Dashboard. Everything Visible.
The browser dashboard (localhost:3000) shows everything in real time:
- Live terminals for every active project
- All client messages in one unified inbox — reply from the dashboard
- Social media drafts waiting for your approval
- Invoices waiting for your CONFIRM
- Cost estimates pending approval
- Background agents running silently
- Token meter — live context usage per project
- Per-project health scores
- Full system logs, screenshots, Playwright videos
Shareable client links too. Token-based, 30-day expiry. Clients see their project status only — nothing else.
The Meta Strategy — AgentJr Builds Itself
After AgentJr launches open source, the plan is to use AgentJr itself to build the next three products:
- AgentPro — Freelancer AI for agencies
- Education Platform — AI-driven learning management
- DevPulse AI — Developer performance monitoring
AgentJr building AgentPro in public — that's a story nobody else can tell. Living proof the tool works. That's the viral moment.
Free. Open Source. Forever.
AgentJr is MIT licensed. No subscription. No usage limits. No cloud dependency. No paid tier hiding features. Everything ships in the open source release.
You run it on your machine. Your code stays on your machine. Your clients talk to your instance. Your vault stays encrypted on your drive.
This is what developer tools should be.
Current Status
AgentJr is in active development. Core systems are being built and tested rigorously.
The plan is to test it thoroughly on real freelance projects for at least one month before open sourcing — because a tool that talks to your clients, touches your codebase, and manages your deployments has to be rock solid before anyone else runs it.
No release date announced. No hype. Just building.
When it's ready — GitHub and npm. Free. Open source.
Follow the Build
- Twitter/X: @hamid_raza_agi — build updates in public
- GitHub: github.com/muhammadhamidraza/agentjr — watch for the release
- Dev.to: Follow me here — architecture deep-dives as each system ships
Built by Hamid Raza — Agentic AI & MERN Stack Developer
AgentJr — Your AI Junior Developer. Always Working.
Top comments (0)