Local-First
Every model call runs against Ollama on-device by default. Your code never has to leave the machine it's on.
Open Source · MIT Licensed
SYJ AI
Autonomous. Local-First. Mobile-Ready.
SYJ AI is a local-first autonomous AI software engineering agent. It plans, codes, reviews, and verifies real files — not chat text — running entirely offline through Ollama. Built mobile-native from Android to Linux, Windows, and macOS, it doesn't assume a laptop, cloud credits, or an always-on connection.
Why SYJ AI
Most AI agents assume a laptop. SYJ AI doesn't.
Most autonomous coding agents are built around a single assumption: a powerful machine, an unmetered cloud API key, and a stable connection. That assumption breaks down for a huge share of the world's developers — on the road, on a phone, on a metered connection, or simply working somewhere the cloud isn't an option. SYJ AI was built from the ground up for that reality: local models, a sandboxed workspace, and a workflow that runs the same whether it's on a laptop or in a Termux session on an Android phone.
Features
Built for offline, sandboxed, real-world engineering work
Offline-Capable
No internet, no problem. Plan, code, and review with zero network dependency once models are pulled.
Modular Architecture
Planning, coding, review, and verification are separable stages — swap models or steps without rewriting the agent.
Real Files, Not Chat Text
SYJ AI edits and writes actual files in your workspace — not code blocks you copy-paste by hand.
Mobile-Native
Developed and tested inside Termux on Android from day one, not ported after the fact.
Safety-Conscious
A sandboxed workspace and a gated shell mean the agent asks before it does anything destructive.
Optional Remote Fallback
Local by default, but you can configure a remote model as a fallback when you want extra headroom.
Fully Tested
A pytest suite and GitHub Actions CI run on every change, not just on release day.
How It Works
An 8-stage engineering workflow, not a single prompt
- 01
Plan
Break the request into a concrete, ordered set of engineering tasks.
- 02
Research
Read the existing codebase and relevant context before touching anything.
- 03
Design
Sketch the approach and interfaces before writing implementation code.
- 04
Code
Write or edit real files in the sandboxed workspace.
- 05
Review
Re-read the diff critically for correctness, style, and edge cases.
- 06
Verify
Run tests and checks to confirm the change actually works.
- 07
Optimize
Tighten the implementation once correctness is established.
- 08
Document
Leave the change legible for the next person — human or agent.
Model Architecture
Two local models, two jobs
DeepSeek Reasoning
Handles planning, debugging, and architecture review — the stages that call for holding more context and weighing trade-offs.
QwenCoder Coding
Handles generation, refactors, and component-level work — the stages that call for fast, precise code output.
Both models run locally through Ollama. No API keys, no per-token billing, no code leaving the device.
Installation
Get running in a few commands
pkg install python git -y
pip install -e .
ollama pull qwen2.5-coder:7b
ollama pull deepseek-r1:7b
syj-ai --helpgit clone https://github.com/SHalimoosavi/SYJ-AI.git
cd SYJ-AI
pip install -e .
ollama pull qwen2.5-coder:7b
ollama pull deepseek-r1:7b
syj-ai --helpgit clone https://github.com/SHalimoosavi/SYJ-AI.git
cd SYJ-AI
pip install -e .
ollama pull qwen2.5-coder:7b
ollama pull deepseek-r1:7b
syj-ai --helpgit clone https://github.com/SHalimoosavi/SYJ-AI.git
cd SYJ-AI
pip install -e .
ollama pull qwen2.5-coder:7b
ollama pull deepseek-r1:7b
syj-ai --helpExact flags and options may evolve — the README is always the source of truth.
From the SAYANJALI Ecosystem
Part of a bigger picture
Sayanjali Nexus
The parent company: AI products, an independent Layer-1 blockchain, enterprise SaaS, and open-source developer tools.
Explore →SYJ Token
The planned utility token settling on SAYANJALI Blockchain, built for enterprise licensing, AI subscriptions, and developer services.
Explore →SAYANJALI Blockchain
An independent Layer-1 blockchain built from scratch in Python — Proof-of-Work consensus, ECDSA wallets, REST API, and CLI.
Explore →Moosavi
Founder & personal portfolio of Syed Ali Hasan Moosavi — the engineering work behind the SAYANJALI ecosystem.
Explore →FAQ
Common questions
What is SYJ AI?
SYJ AI is an open-source, local-first autonomous AI software engineering agent. It plans, writes, reviews, and verifies code by running local language models through Ollama, so it works fully offline without sending code to the cloud.
Does SYJ AI work without an internet connection?
Yes. SYJ AI is offline-capable by default, running local models such as DeepSeek and QwenCoder through Ollama, with an optional remote fallback if you choose to configure one.
Can SYJ AI run on Android?
Yes. SYJ AI is mobile-native and supports installation on Android through Termux, in addition to Linux, macOS, and Windows.
What models does SYJ AI use?
SYJ AI pairs DeepSeek for reasoning — planning, debugging, architecture review — with QwenCoder for code generation, refactors, and component work. Both run locally through Ollama.
Is SYJ AI free and open source?
Yes. SYJ AI is released under the MIT license and its source is publicly available on GitHub.
Open Source & Community
MIT licensed. Built in the open.
SYJ AI is released under the MIT license and developed in the open. Issues and pull requests are welcome, GitHub Actions runs the test suite on every change, and the project is still an early, honest v0.1.0 — built for real-world use and shaped by community feedback, not a finished product.