SHIMS Omni

Self-Hosted Intelligent Multi-agent System

SHIMS is a local-first, multi-modal AI assistant that runs entirely offline with Ollama, or routes to cloud providers like Gemini, OpenAI, Anthropic, Kimi, DeepSeek, and Qwen. It includes SHIMS Omni (the chat/assistant layer) and the Council of the Wise (multi-agent debate mode).

License: Elastic License 2.0 — free for personal and internal use.

---

What is SHIMS?

SHIMS is designed to be your own private AI operating system:

• Private by default — runs local LLMs through Ollama; cloud keys are optional.
• Multi-agent — specialist agents for coding, research, chemistry, planning, and self-improvement.
• Multi-modal — chat, voice, documents, images, video, code interpreter, and screen/memory ingestion.
• Self-improving — guarded self-evolution loop with validation, approval, and rollback.

---

Included apps

---

Key features

SHIMS Omni

• Local-first chat with Ollama and automatic fallback to tiny models if the chosen model is slow.
• Optional cloud routing for Gemini, OpenAI, Anthropic Claude, Kimi, DeepSeek, Qwen, and local HuggingFace endpoints.
• Voice input/output with browser speech controls.
• Document generation: TXT, DOCX, PDF, XLSX, PPTX.
• Image and video generation with local renderers and provider hooks.
• In-chat Python code interpreter with matplotlib capture.
• Vector memory and retrieval (omni-brain) for durable facts and source ingestion.
• Long-horizon task planner with dependency-aware wave execution.
• Guarded self-evolution: propose → validate → approve → apply → archive.
• Desktop automation, mail assistant, browser agent, and scheduler.

Council of the Wise

• Brings up to five agents into the same conversation: Omni, Gemini, Claude, GPT, and Factory.
• Each agent responds in parallel; a Chair synthesizes a final answer and action plan.
• The Chair can invoke any SHIMS tool — shell commands, file edits, plans, schedules, browser automation, mail, App Factory, self-evolution patches, and more.
• Supports free chat, structured debate, and cross-instance collaboration between a primary SHIMS instance and an isolated Local Factory instance.

---

Quick start (Windows)

1. Install Python 3.11 or 3.12.
2. Install Git.
3. Install Ollama.
4. Install FFmpeg and add it to PATH.
5. Install or update NVIDIA drivers (for local GPU acceleration).
6. Clone this repo to a short path like C:\SHIMS.
7. Copy .env.example to .env and fill in your API keys (optional).
8. Run install_windows.bat.
9. Run START_SHIMS.bat to launch the Desktop Bridge + Omni.

Open Omni at http://127.0.0.1:8010.

Recommended first model

ollama pull qwen2.5:7b

For lower VRAM:

ollama pull llama3.2:3b

---

Configuration

Copy the example environment file and edit it:

copy .env.example .env

Key variables:

Never commit .env — it is already ignored.

---

Development

Run tests:

.venv/Scripts/python -m pytest tests/ -v

Start the backend directly:

.venv/Scripts/python -m uvicorn backend.app.main:app --host 127.0.0.1 --port 8010

---

Architecture

• backend/app/main.py — FastAPI core for Omni chat, model routing, and API surface.
• shared/ — Agent tools, wave engine, memory, planner, vision, code interpreter, DuoBot, and bridge.
• frontend/ — Vanilla JS/HTML/CSS frontend for Omni.
• apps/ — Vertical apps generated by the SHIMS App Factory (e.g., Todo Demo, Sheena AI Wellness).
• tests/ — pytest regression and feature tests.

---

Safety and limitations

This is a functional local-first foundation. Validate workflows, backup policy, access control, and data integrity before any regulated or production use.

---

License

This project is licensed under the Elastic License 2.0.

You are free to use, modify, and distribute SHIMS for non-commercial and internal purposes. Offering SHIMS as a hosted or managed service, or selling it as a product, requires a separate commercial license from the copyright owner.

---

Support

• Open an issue in this repo.
• Start a discussion for feature requests or questions.