Sessions-AI — User Documentation
Overview
Sessions-AI is a private AI chat app that runs entirely on your iPhone, iPad, or Mac. No cloud servers, no account required—just choose a model, start a conversation, and keep full control of your data.
Key Features
- On-device language models—your conversations never leave your device
- Multiple chats stored as markdown files in the Files app
- Knowledge base to ground answers in your own documents
- Personas with per-persona system prompts and knowledge bases
- Built-in tools: calculator, web search, calendar, reminders, and more
- Optional MCP integration for third-party tools
- No telemetry, no analytics, no account
Getting Started
First Launch
When you first open Sessions-AI:
- Choose a model — Open Settings and select a model from the catalog. If it isn't downloaded yet, tap to download it. Start with a smaller model for quick responses, or a larger one for complex tasks.
- Start a chat — Create a new chat from the sidebar (or it may open automatically). Type your message and send.
- Wait for generation — The model runs on your device. First responses may take a moment as the model loads and processes.
Note: No account or sign-in is required. All processing happens locally on your device.
Creating and Managing Chats
Each conversation is a separate chat. Chats are saved as markdown files on your device:
- iOS: In the app's Documents folder (visible in the Files app)
- macOS: In Documents/Sessions-AI
You can create as many chats as you like. Start a new chat for a fresh topic, or pick up an existing one anytime from the sidebar.
Selecting a Model
In Settings > Model:
- Browse the model catalog (powered by Hugging Face)
- Select a model—the app will download it if needed
- Choose models that fit your device: smaller models run faster; larger models handle more complex tasks but need more storage and memory
- Remove models you no longer need to free space
Chat and Conversations
Sending Messages and Attachments
Type your message in the input field and send. You can attach files before sending to give the assistant context. Supported formats include text files, markdown, PDFs, and more. The assistant can read attached files to answer questions or summarize content.
Regenerating and Canceling
- Regenerate — If you're not satisfied with the last assistant reply, use the Regenerate button to have the model try again with the same user message.
- Cancel — While the model is generating, tap Cancel to stop. The partial reply is discarded; you can edit and resend if you like.
Chat Storage and Files App
Chats are stored as plain markdown (.md) files. On iOS, they appear in the Files app under the app's Documents. On Mac, they're in Documents/Sessions-AI. You can open, copy, or back up these files with standard tools.
Knowledge Base
The knowledge base lets you add your own documents so the assistant can search them and ground its answers in your content—all on device. No documents are uploaded to the cloud.
Adding Documents
In Settings > Knowledge Base (within the active persona):
- Tap Add or drag documents into the knowledge base section
- Supported formats include text, markdown, and PDF
- Documents are chunked and indexed locally using full-text search (FTS5)
How RAG Works
When you ask a question and the knowledge base has documents:
- The app searches your documents locally for relevant chunks
- Matching content is injected into the system prompt as context
- The model uses that context to answer—without ever sending your documents anywhere
Last Retrieval / Citations
Settings may show "Last retrieval" with the titles of documents that were used for the most recent answer. This helps you see which of your documents informed the response.
Note: The knowledge base is per-persona. Each persona has its own set of documents.
Personas
Personas let you create different "assistant identities" with separate system prompts, knowledge bases, and MCP server configs.
What Personas Provide
- System prompt — Custom instructions that shape how the assistant responds
- Knowledge base — Each persona has its own document set
- MCP servers — Persona-specific MCP server URLs for tools
Creating and Switching Personas
In Settings > Personas:
- Add a new persona and give it a name
- Edit the system prompt and add documents as needed
- Switch the active persona to change which one is used for new chats
Built-in Tools
The assistant can use various tools to help you. Ask in natural language—for example, "What's 15% of 240?" or "Add a reminder to call Mom tomorrow."
Calculator and Unit Conversion
Ask math questions or request unit conversions (e.g., "Convert 5 miles to km" or "100 Fahrenheit to Celsius").
Calendar
List events, add events, or delete events. Example: "What's on my calendar for the next 7 days?" or "Add a calendar event: Team standup tomorrow 9am to 9:30am." Uses Calendar access on your device—data stays local.
Reminders
List, add, complete, or delete reminders. Example: "Add a reminder: Buy milk" or "List my reminders." Uses Reminders on your device.
Optional Web Search
When enabled in Settings, the assistant can search the web via DuckDuckGo. Only the search query is sent—not your full conversation. Enable or disable in Settings.
Fetch Page and Wikipedia
The assistant can fetch content from a URL or look up Wikipedia articles when you ask for information from the web.
Clipboard, Contacts, Maps
Read or write clipboard content, search contacts, open Maps or get directions—all on device.
Planning
For multi-step tasks, the assistant can create a plan, show the current plan, and update step status. Useful for complex projects.
Platform-Specific Tools
macOS only: File search (Spotlight), Mail (draft emails), Notes (create Apple Notes), Screen capture, and System control (open apps) are available on Mac. On iOS, these tools are not offered.
MCP (Model Context Protocol)
MCP lets you connect third-party servers that expose tools the assistant can call. For example, a server might provide a "database query" or "code execution" tool.
Adding MCP Servers
In Settings > MCP (within the active persona), add MCP server URLs. The app connects via HTTP and registers the tools from each server. MCP config is per-persona.
Third-party disclaimer: MCP servers you add are third parties over which we have no control. Their privacy practices, data handling, and security are their responsibility. We recommend reviewing the documentation and privacy policy of any MCP server before adding it.
Settings
Model Selection
Choose which language model to use. Models are downloaded from the in-app catalog and stored on your device.
Generation Parameters
Adjust how the model responds:
- Max tokens — Maximum length of each reply
- Temperature — Higher values increase creativity; lower values make responses more focused
Advanced Sampling
Power users can fine-tune sampling (top-k, top-p, min-p, typical-p, repetition penalty, mirostat, etc.) in the Advanced sampling section.
Web Search Toggle
Enable or disable web search. When disabled, the assistant will not use DuckDuckGo to search the web.
Privacy Summary
Settings includes a privacy summary explaining that Sessions-AI runs locally, chats stay on device, and web search (when used) sends only the query.
Privacy and Data
Complete Privacy
All processing runs on your device. Chats, the knowledge base, and model data stay on your iPhone, iPad, or Mac.
- No cloud inference
- No user accounts
- No telemetry or analytics
- Optional web search sends only the search query (DuckDuckGo), not your full conversation
- Calendar and reminder access are used only on device
If you add MCP servers, those servers are third parties. We have no control over their data handling. See the Privacy Policy for full details.