SysSage — Local AI System Assistant (Windows)

SysSage is a privacy-first, on-device assistant for Windows that uses local LLM adapters and OS tooling to answer questions about your system, run diagnostics, automate workflows, and assist with file/process/hardware queries — all while keeping data on your machine.

Table of Contents

• Why SysSage
• Proposed solution
• Key features
• Quickstart (Windows / PowerShell)
• Project layout
• Architecture (high level)
• Security & Privacy
• Extending SysSage
• Development notes
• Troubleshooting & common notes
• Contributing & roadmap
• License
• Contact

Why SysSage

• System troubleshooting, inventory, and personal assistance often require multiple tools, admin rights, or cloud services that expose sensitive data.
• Developers, power users, and administrators need fast, contextual, and private help about processes, running services, hardware, and files without sending data to the cloud.

Proposed solution

SysSage provides a modular agent framework and a Streamlit UI that connects a suite of OS-level tools (process, file, hardware, browser artifacts) with local LLMs (via llm/ollama_wrapper.py or other adapters). It focuses on local-first intelligence, scripted automation, and extensibility.

Key features

• Privacy-first local LLM integration — use Ollama (or swap in another local/secure model) so prompts and context can stay on-device.
• System inspection tools — process, hardware, file, and browser helpers under tools/ (e.g., psutil, pywin32).
• Agent-based automation — small agents in agents/ can monitor, execute, and report tasks (executor, monitor).
• Crew runner / Orchestration — crew_runner.py coordinates multiple agents for multi-step workflows.
• Simple Streamlit UI — ui/streamlit_ui.py plus an intent dispatcher (ui/intent_dispatcher.py) for quick interactions and saved histories.
• Scriptable experiments — assistant.py, simple_approach.py provide examples and starting points for new features.

Quickstart (Windows / PowerShell)

1. Create and activate a virtual environment

python -m venv .venv
.\.venv\Scripts\Activate.ps1

2. Install dependencies

pip install -r requirements.txt

3. (Optional) Configure local LLMs

• If you plan to use Ollama or another local LLM, install and configure it separately. See llm/ollama_wrapper.py for integration points.

4. Run the Streamlit UI

streamlit run .\ui\streamlit_ui.py

5. Run the agent crew example

python .\crew_runner.py

Project layout

• agents/ — agent implementations (e.g., executor_agent.py, monitor_agent.py).
• llm/ — local LLM adapter(s) and wrappers.
• tasks/ — task orchestration helpers.
• tools/ — OS and utility tooling (process, file, hardware, browsing helpers).
• ui/ — Streamlit UI, intent dispatcher, and front-end wiring.
• crew_runner.py — example orchestrator that launches agent crews.
• assistant.py, simple_approach.py — experimental entrypoints and scripts.
• requirements.txt — pinned Python dependencies.

Architecture (high level)

1. UI (Streamlit)

   • Collects intents and displays agent outputs. Sends structured requests to the intent dispatcher.

2. Intent Dispatcher (ui/intent_dispatcher.py)

   • Maps UI intents to agent/task workflows and routes them to the crew runner or direct tools.

3. Agents & Crew Runner

   • Agents encapsulate capabilities (monitoring, execution, file analysis). crew_runner.py composes agents to run multi-step flows.

4. Tools

   • Deterministic helpers that perform system actions or data retrieval (psutil wrappers, process scanning, file search, browser artifact parsing).

5. LLM Adapter

   • llm/ollama_wrapper.py shows how to place contextual prompts and combine deterministic outputs with model reasoning while retaining privacy.

Security & Privacy

• Local-first operation: LLM prompts and context can stay local when using on-device models.
• Minimal telemetry: default behavior is to avoid external logging of sensitive outputs; add explicit opt-in if remote logging is required.
• Elevated actions: some tools require admin rights — the UI will surface warnings and request appropriate permissions.

Extending SysSage

• Add a new tool: create a helper in tools/, write a small wrapper, and register it with an agent.
• Add a new agent: follow the pattern in agents/ and expose a simple interface (start/stop/handle_intent).
• Swap LLM provider: implement a new adapter under llm/ that matches the wrapper interface.

Development notes

• Recommended Python: 3.10+.
• Tests: add unit tests for tool logic (avoid calling LLMs in unit tests; mock adapters).
• CI: small lint and test step (GitHub Actions recommended).

Troubleshooting & common notes

• Streamlit launching issues: ensure your venv is active and streamlit is installed.
• Missing OS-level permissions: run PowerShell as Administrator for tasks that require system-level access.
• LLM adapter errors: confirm the local model server is running and the wrapper configuration matches your installation.

Contributing & roadmap

• Open issues to suggest features or report bugs.
• Short-term roadmap: improve agent orchestration, add richer browser artifact parsing, and include more robust offline prompt templates.

SysSage — Local AI System Assistant (Windows)

SysSage is a privacy-first, on-device assistant for Windows that uses local LLM adapters and OS tooling to answer questions about your system, run diagnostics, automate workflows, and assist with file/process/hardware queries — all while keeping data on your machine.

Why SysSage exists (Problem)

• System troubleshooting, inventory, and personal assistance often require multiple tools, admin rights, or cloud services that expose sensitive data.
• Developers, power users, and administrators need fast, contextual, and private help about processes, running services, hardware, and files without sending data to the cloud.

Proposed solution

• SysSage provides a modular agent framework and a Streamlit UI that connects a suite of OS-level tools (process, file, hardware, browser artifacts) with local LLMs (via ollama_wrapper.py or other adapters). It focuses on local-first intelligence, scripted automation, and extensibility.

Key features

• Privacy-first local LLM integration: use Ollama (or swap in another local/secure model) so prompts and context can stay on-device.
• System inspection tools: process, hardware, file, and browser helpers under tools/ (e.g., psutil, pywin32 usages).
• Agent-based automation: small agents live in agents/ and can monitor, execute, and report tasks (executor, monitor).
• Crew runner / Orchestration: crew_runner.py coordinates multiple agents for multi-step workflows.
• Simple Streamlit UI: ui/streamlit_ui.py plus an intent dispatcher (ui/intent_dispatcher.py) for quick interactions and saved histories.
• Scriptable experiments: assistant.py, simple_approach.py provide examples and starting points for new features.

Short developer quickstart (Windows / PowerShell)

1. Create and activate a virtual environment

python -m venv .venv
.\.venv\Scripts\Activate.ps1

2. Install dependencies

pip install -r requirements.txt

3. (Optional) Configure local LLMs

• If you plan to use Ollama or another local LLM, install and configure it separately. See llm/ollama_wrapper.py for integration points.

4. Run the Streamlit UI

streamlit run .\ui\streamlit_ui.py

5. Run the agent crew example

python .\crew_runner.py

Project layout (what's where)

• agents/ — agent implementations (e.g., executor_agent.py, monitor_agent.py).
• llm/ — local LLM adapter(s) and wrappers.
• tasks/ — task orchestration helpers.
• tools/ — OS and utility tooling (process, file, hardware, browsing helpers).
• ui/ — Streamlit UI, intent dispatcher, and front-end wiring.
• crew_runner.py — example orchestrator that launches agent crews.
• assistant.py, simple_approach.py — experimental entrypoints and scripts.
• requirements.txt — pinned Python dependencies.

Architecture (high level)

1. UI (Streamlit)

   • Collects intents and displays agent outputs. Sends structured requests to the intent dispatcher.

2. Intent Dispatcher (ui/intent_dispatcher.py)

   • Maps UI intents to agent/task workflows and routes them to the crew runner or direct tools.

3. Agents & Crew Runner

   • Agents encapsulate capabilities (monitoring, execution, file analysis). crew_runner.py composes agents to run multi-step flows.

4. Tools

   • Deterministic helpers that perform system actions or data retrieval (psutil wrappers, process scanning, file search, browser artifact parsing).

5. LLM Adapter

   • llm/ollama_wrapper.py shows how to place contextual prompts and combine deterministic outputs with model reasoning while retaining privacy.

Security & Privacy

• Local-first operation: LLM prompts and context can stay local when using on-device models.
• Minimal telemetry: default behavior is to avoid external logging of sensitive outputs; add explicit opt-in if remote logging is required.
• Elevated actions: some tools require admin rights — the UI will surface warnings and request appropriate permissions.

Extending SysSage

• Add a new tool: create a helper in tools/, write a small wrapper, and register it with an agent.
• Add a new agent: follow the pattern in agents/ and expose a simple interface (start/stop/handle_intent).
• Swap LLM provider: implement a new adapter under llm/ that matches the wrapper interface.

Development notes

• Recommended Python: 3.10+.
• Tests: add unit tests for tool logic (avoid calling LLMs in unit tests; mock adapters).
• CI: small lint and test step (GitHub Actions recommended).

Troubleshooting & common notes

• Streamlit launching issues: ensure your venv is active and streamlit is installed.
• Missing OS-level permissions: run PowerShell as Administrator for tasks that require system-level access.
• LLM adapter errors: confirm the local model server is running and the wrapper configuration matches your installation.

Contributing & roadmap

• Open issues to suggest features or report bugs.
• Short-term roadmap: improve agent orchestration, add richer browser artifact parsing, and include more robust offline prompt templates.

License

• Add a LICENSE to the repository if you plan to open-source. MIT or Apache-2.0 are common choices.

Contact

• Use the repository issue tracker for collaboration and questions.

— If you want this written in a slightly shorter pitch format, converted to a slide-friendly one-pager, or saved directly into the repo as README.md, tell me and I will update the file or create the alternate formats.