Reiseki — Local-first AI assistant for files, reminders, and document generation

Reiseki runs inference locally and does not require a cloud LLM API. It provides a FastAPI-based chat UI, optional desktop packaging via pywebview, local SQLite memory, and tool access for files, reminders, documents, and lightweight data analysis.

---

---

Features

• Local LLM inference via Ollama
• File-system tools: list, read, write, create directories
• PDF support: read and extract text from PDFs, generate PDF documents
• Local memory persisted in SQLite — you control what the agent remembers
• Conversation history across sessions — the agent recalls previous conversations
• Reminders / appointments with toast notifications in the UI
• Document export: .docx, .xlsx, .csv, .pdf
• Basic data analysis and chart generation
• Streaming responses via SSE
• Smartphone access via QR code toggle in the UI
• Desktop launcher via pywebview
• Model switcher — change the Ollama model directly in the UI without restarting

---

Privacy / Security model

Local by default

• Ollama inference runs locally
• Memory is stored locally in SQLite
• File access is restricted to AGENT_ROOT
• Existing files cannot be overwritten or deleted by the agent

Network use

• Installing dependencies
• Pulling Ollama models
• Optional LAN access if enabled

Caution

Reiseki restricts file access to AGENT_ROOT, but it is not a sandbox or security boundary. For sensitive use cases, run it in a dedicated workspace, VM, or container.

LAN access: When the smartphone toggle is enabled, any device on the same network can reach the agent without authentication. Only enable this on a trusted private network (e.g. your home Wi-Fi). Never use it on public or shared networks such as hotels, cafés, or offices.

Important

AGENT_ROOT defines the workspace the agent may access.
For safer use, run Reiseki inside a dedicated workspace or set AGENT_ROOT explicitly.

---

Requirements

• Python 3.10+
• Ollama

Pull a model first:

ollama pull gemma4:e2b

Install

Option 1: Release installer

Ollama must be installed before running the installer.

Download the latest release from the Releases page:

• ReisekiSetup.exe — Windows
• Reiseki.dmg — macOS

---

Option 2: From source

git clone https://github.com/Flo1632/reiseki.git
cd reiseki
pip install -r requirements.txt

---

Option 3: Script installer

For systems with Python and Ollama already installed:

• install.sh — macOS / Linux
• install.bat — Windows

These installer scripts create a virtual environment, install dependencies, pull the default model, and generate a launch script.

---

Run

Browser

python agent/agent.py

then open http://localhost:8000

Desktop window

python agent/launcher.py

---

Configuration

Variable	Default	Description
AGENT_MODEL	gemma4:e2b	Ollama model
AGENT_ROOT	~/Reiseki	Accessible workspace root

AGENT_MODEL=gemma4:e2b AGENT_ROOT=~/documents python agent/agent.py

---

Example prompts

• Summarize all Markdown files in this folder.
• Create a todo.xlsx with task, deadline, and status columns.
• Read notes.pdf and extract the main points.
• Search the web for the latest FastAPI release notes.
• Remind me tomorrow at 09:00 to send the invoice.

---

Tools

Tool	Description
list_directory	List files and folders
read_file	Read file content (text, PDF)
write_file	Create a new file
create_directory	Create a directory
save_memory	Store memory in SQLite
list_memories	List saved memories
add_appointment	Create a reminder
list_appointments	List reminders
create_docx	Generate a Word file
create_pdf	Generate a PDF file
create_xlsx	Generate an Excel file
create_csv	Generate a CSV file
analyse_data	Analyze tabular data
create_chart	Render a matplotlib chart

---

Stack

Layer	Technology
LLM backend	Ollama (Gemma 4)
API	FastAPI + Uvicorn
Validation	Pydantic v2
Memory	SQLite
Desktop wrapper	pywebview
Frontend	HTML / CSS / JavaScript

---

Disclaimer

Caution

Reiseki is provided "as is", without warranty of any kind, express or implied. Use it at your own risk. The authors are not liable for any claim, damages, or other liability arising from the use of the software, including but not limited to data loss, incorrect outputs, failed automations, or system misconfiguration. Reiseki can read and modify files inside AGENT_ROOT. Always review outputs and keep backups of important data. Reiseki is not a security boundary or sandbox. If you need stronger isolation, run it inside a dedicated workspace, VM, or container. Reiseki uses a local large language model (LLM) via Ollama to generate responses, files, and analyses. Additionally, please keep the following in mind:

• No guarantee of correctness — LLM outputs can be inaccurate, incomplete, or entirely fabricated ("hallucinations"). Always review generated content before relying on it.
• File operations at your own risk — The agent can read, write, and overwrite files within AGENT_ROOT. Make sure you have backups of important data.
• Not professional advice — Generated documents, analyses, and scheduling suggestions do not replace professional legal, financial, or medical advice.
• Model-dependent quality — Output quality depends entirely on the Ollama model you choose. Reiseki itself is not an AI model.
• Built with AI — This project was built entirely with the tool Claude Code by Anthropic.

---

License

MIT — see LICENSE