AI accelerates engineering without removing human control.
Gebeta Sovereign Code Assistant is a local-first AI engineering environment that enables developers and teams to code, review, refactor, test, and execute agent workflows without exposing proprietary source code to third-party AI providers.
It combines local LLM inference (Ollama), IDE-native agent workflows (Continue), and a control layer of policies, approvals, and guardrails into a practical system for sovereign software development.
- Local First – Models run locally, code stays local
- Human Approval – Sensitive actions require explicit review
- Controlled Agents – AI operates inside policy boundaries
- Auditability – Engineering actions are reviewable
- Team Standardization – Repeatable, governed AI workflows
| Feature | Description |
|---|---|
| 🔒 Local Inference | Run coding models locally using Ollama — no cloud dependency |
| 🤖 Agent-Ready | Multi-step coding workflows with Continue |
| ✅ Human Approval | Explicit review before sensitive actions |
| 🛡️ Project Guardrails | Repo-specific rules and coding standards |
| 📋 Auditability | Local history, approvals, and action logs |
| 👥 Team Standardization | Shared configs and deployment patterns |
| Zero external connectivity required | |
| 🎛️ Two Deployment Modes | Maximum Privacy or Productivity Mode |
Get up and running in 10 minutes.
git clone https://github.com/gebetasuq/gebeta-Sovereign-code-assistant-
cd gebeta-Sovereign-code-assistant-- Install Ollama
# macOS/Linux
curl -fsSL https://ollama.com/install.sh | sh
# Windows: Download from https://ollama.com- Start Ollama & Pull Models
# Start the Ollama server (keep this terminal open)
ollama serve
# In a new terminal, pull recommended models
ollama pull qwen2.5-coder:7b
ollama pull codellama:7b- Install VS Code & Continue
· Install Visual Studio Code · Install the Continue extension
- Configure Continue
# macOS/Linux
mkdir -p ~/.continue
cp configs/continue-config.yaml ~/.continue/config.yaml
# Windows
mkdir %USERPROFILE%\.continue
copy configs\continue-config.yaml %USERPROFILE%\.continue\config.yaml- Start Coding
Open VS Code, open the Continue sidebar (Cmd+Shift+P → "Continue: Open Chat"), and start coding with AI assistance!
Best for: Fintech, proprietary IP, compliance-sensitive environments
Stack:
- Ollama (local inference)
- Continue (IDE agent)
- VS Code (telemetry minimized)
- Local terminal only
Setup:
cp configs/continue-config-safe.yaml ~/.continue/config.yamlBest for: Multi-agent workflows, faster execution
Stack:
· Ollama + Continue · Warp terminal (with ZDR enabled) · Hardened internet-enabled environment
Setup:
# Install Warp (macOS)
brew install --cask warp
# Enable Zero Data Retention in Warp settings
cp configs/continue-config.yaml ~/.continue/config.yaml| Use Case | Model | RAM | Command |
|---|---|---|---|
| Fast autocomplete | qwen2.5-coder:1.5b |
~2 GB | ollama pull qwen2.5-coder:1.5b |
| Balanced coding agent | qwen2.5-coder:7b |
~6 GB | ollama pull qwen2.5-coder:7b |
| General assistant + docs | llama3.1:8b |
~8 GB | ollama pull llama3.1:8b |
| Low RAM fallback | phi3:mini |
~2.5 GB | ollama pull phi3:mini |
| Risk | Mitigation |
|---|---|
| Third-party AI training on code | Local models only |
| Cloud prompt retention | No data sent to hosted providers |
| Accidental source leakage | Air-gappable configuration |
| Over-permissioned agents | Manual approval required |
| Silent execution | Agent asks before running commands |
- Malicious local dependencies (npm, pip, etc.)
- Insecure commands approved by the user
- Compromised OS / endpoint malware
- Package manager supply-chain attacks
- Git remote misconfiguration
- Insider threats
Important: This is a control-first system, not a convenience-first system. Human review is always required.
Authentication endpoints are protected against brute‑force attacks:
- Login: 10 requests per minute per IP
- Registration: 5 requests per 5 minutes per IP
Exceeding the limit returns HTTP 429 with a Retry-After header. Rate limiting is implemented in both FastAPI and Spring Boot templates.
| Document | Description |
|---|---|
| QUICKSTART.md | Get started in 10 minutes |
| SECURITY_AND_TRUST.md | Threat model and trust boundaries |
| TEAM_DEPLOYMENT.md | Scale to your team |
| USE_CASES.md | Real-world examples |
| WHY_GEBETA.md | Founder vision and philosophy |
| ROADMAP.md | Product roadmap |
| CONTRIBUTING.md | How to contribute |
| TESTING.md | Running tests and coverage reports |
| PERFORMANCE_TUNING.md | Memory limits, JVM tuning, hardware adjustments |
Gebeta Sovereign Code Assistant is already being used to build the Global AI Civilization Platform (GACP) – a sovereign, multi‑agent AI operating architecture.
GACP includes:
- 10 specialised AI agents (Orchestrator, Builder, Guardian, Mitu AI, etc.)
- 8 architecture layers (from user interface to global federation)
- Self‑improving learning loops
- Economic resource management
- Sovereign and federation layers
If Gebeta can help build GACP, it can help build your sovereign AI systems too.
gebeta-sovereign-code-assistant/
│
├── README.md # This file
├── LICENSE # MIT License
├── QUICKSTART.md # Quick start guide
├── SECURITY_AND_TRUST.md # Security documentation
├── TEAM_DEPLOYMENT.md # Team setup guide
├── USE_CASES.md # Real-world examples
├── WHY_GEBETA.md # Founder vision
├── ROADMAP.md # Product roadmap
├── CONTRIBUTING.md # Contribution guidelines
├── TESTING.md # Test execution guide
├── PERFORMANCE_TUNING.md # Performance optimization
├── BUILT_WITH_GEBETA_SOVEREIGN_CODE_ASSISTANT.md # Built With story (GACP)
├── .gitignore # Git ignore rules
├── .gitleaks.toml # Secret scanning configuration
├── docker-compose.yml # Full-stack deployment
│
├── .github/
│ ├── ISSUE_TEMPLATE/ # Issue templates
│ └── workflows/ # CI test workflows
│
├── configs/ # Ready-to-use configurations
│ ├── continue-config.yaml
│ ├── continue-config-safe.yaml
│ ├── continue-config-team.yaml
│ ├── continue-config-lowram.yaml
│ ├── gebeta-rules.md
│ └── safe-command-policy.md
│
├── docs/ # Additional documentation
│ ├── architecture.md
│ └── deployment-modes.md
│
├── examples/ # Example workflows
│ └── example-agent-prompts.md
│
└── templates/ # Starter templates
├── fastapi-service-template/
├── react-frontend-template/
└── springboot-service-template/
- Local models may be slower than cloud models
- Agent tool use varies by model and hardware
- Large repositories may require context tuning
- Code quality depends on model choice and human review
- Some actions need repeated approval
- Autocomplete quality may be below premium hosted tools
Positioning: Gebeta promises controlled AI, not perfect AI. That is a stronger promise.
| Version | Focus | Timeline |
|---|---|---|
| V1 | Foundation — Documentation, configs, starter kit | ✅ April 2026 |
| V1.1 | Security & Stability — Rate limiting, token refresh, error handling | ✅ April 2026 |
| V2 | Platform — Web portal, onboarding, analytics | Q3 2026 |
| V3 | Enterprise — Team control plane, governance, audit dashboard | Q1 2027 |
If Gebeta Sovereign Code Assistant helps you build with control and privacy, please star this repository and share it with your team.
We welcome contributions! Please see CONTRIBUTING.md for guidelines.
Ways to contribute:
- 🐛 Report bugs
- 💡 Suggest features
- 📝 Improve documentation
- 🔧 Submit code improvements
- 📢 Share with your network
This project is licensed under the MIT License — see LICENSE for details.
Mohammed B. Kemal
Founder & System Architect, Gebeta Universe
- 🌐 Website: https://gebetauae.com
- 🔗 LinkedIn: https://www.linkedin.com/in/mohammed-b-kemal
- 🐦 Twitter :https://x.com/mickyMi08136043
- Ollama — Local LLM runtime
- Continue — IDE AI assistant
- VS Code — Code editor
- Warp — Modern terminal (optional)
Built with ❤️ for sovereign engineering.
Document version: 1.0.1 | Last updated: April 2026
© 2026 Gebeta Universe. All rights reserved.