User Guide

Getting started with Obsidia

A practical guide to using the AI assistant — what it can do, how to get the best from it, and what to keep in mind as you work.

Section 01

How AI thinks

Before diving in, it helps to understand a few key ideas about how AI actually works. You don't need a technical background — these six concepts give you enough to work with AI confidently and get consistent results.

01
It predicts, not retrieves
The AI generates responses by predicting the most likely continuation of your input — it doesn't look things up in a database. That means it can produce plausible-sounding but incorrect answers, especially for specific facts, dates, or citations.
In practice: Always verify specific facts, case references, or statistics before relying on them.
02
Context is everything
The AI responds to what's in the current conversation window. It doesn't remember previous chats, and it doesn't know anything about you unless you tell it. The more context you give — your role, the purpose of the task, the relevant background — the better the output.
In practice: Start messages with who you are and what you need, especially for professional tasks.
03
Temperature controls creativity
Temperature is a setting that controls how varied or predictable the AI's responses are. Low temperature means more focused, consistent answers — better for analysis, legal drafting, and compliance work. Higher temperature suits brainstorming and creative tasks.
In practice: For most professional tasks, use a low or medium setting. Adjust in your profile settings.
04
Long documents get compressed
There's a limit to how much text the AI can hold in context at once. For very long documents or conversations, earlier content may be summarised or dropped. This is why breaking large tasks into steps produces better results than asking for everything at once.
In practice: For long documents, ask the AI to focus on a specific section rather than summarising the whole thing in one go.
05
It can't browse the internet
The AI draws on its training data and the documents in your workspace — it cannot access the web, check live databases, or retrieve real-time information. For current case law, recent regulatory updates, or live market data, you'll need to provide those sources directly.
In practice: Paste in the relevant text or upload the document — don't assume the AI already has it.
06
Prompting is a skill
The quality of the AI's response is directly tied to the quality of your request. Vague prompts produce vague answers. Specific, well-structured prompts with clear objectives consistently produce better output — this improves quickly with practice.
In practice: See Section 03 for a short list of techniques that make a measurable difference.
Section 02

Using the application

A walkthrough of the key things you'll do in Obsidia. The application is designed to feel familiar — it works like a private, secure version of the AI chat tools you may have used before.

1
Signing in
Sign in using the method your organisation has configured — either your work email and password, or your Microsoft account via Azure AD single sign-on. If you haven't received an invitation, contact your administrator.
Go to your organisation's login URL → enter your credentials
2
Choosing a workspace
Workspaces are separate areas within the application — each contains its own documents and is scoped to a team or function (for example: Legal, HR, Compliance). You'll only see the workspaces you've been given access to. Select the one that's relevant to what you're working on.
Sidebar → select workspace from the list
3
Starting a conversation
Type your question or request in the chat box and press Enter. The AI will respond based on the documents and context available in the current workspace. Each conversation is independent — starting a new chat gives the AI a fresh context window.
Chat → type your message → press Enter or click Send
4
Uploading a document to a conversation
You can attach a document directly to a conversation — useful when you want the AI to analyse, summarise, or draft based on a specific file. Supported formats include PDF, Word documents, and plain text. The document is available for that conversation only and is not stored in the workspace library.
Chat → paperclip icon → select file → send with your message
5
Viewing your conversation history
Your past conversations are saved and accessible from the sidebar. You can pick up where you left off, or use a previous conversation as a reference. If a conversation was very long, the AI starts fresh when you open it — it doesn't re-read the entire history automatically.
Sidebar → Conversations → select a past conversation
6
Exporting a conversation
You can export any conversation as a Markdown file or as a PDF — useful for saving a piece of work, sharing output with a colleague, or keeping a record of an AI-assisted draft. The export includes the full conversation thread.
Open conversation → ⋮ menu (top right) → Export as Markdown or Export as PDF
7
Setting your temperature preference
You can set your preferred temperature in Settings — this controls how creative or precise the AI's responses are across all your queries. For most professional tasks, a low or medium setting is recommended. See Section 01, Concept 03 for guidance on which setting suits different task types.
Your name (bottom left) → Settings → Preferences → Temperature
8
Using a persona
A persona is a set of instructions that tells the AI how to behave — for example, to respond like a legal analyst, or to keep answers concise and formal. Your administrator assigns personas to workspaces. You can set your preferred persona in Settings, which applies to all your queries across the application.
Your name (bottom left) → Settings → Personas → select your preferred persona
9
Viewing your query history
Every query you send to the AI is logged for governance and compliance purposes. You can view your own record at any time from the My Activity page — it shows a timestamped list of your queries, the workspace they were sent in, the response type, and a brief preview of the query and response. You can filter by date range and workspace, and export your records as CSV or JSON.
Sidebar → My Activity
Privacy notice. Only a short preview of each query is displayed in this view — full query content is not shown here. Your organisation's administrator may have visibility of query metadata (previews, timestamps, token counts) as part of their governance responsibilities. This logging is a platform requirement under your organisation's data governance policy.
Section 03

Getting better results

These seven habits make a measurable difference to the quality of AI output. None of them require technical knowledge — they're about communicating clearly with a system that takes your words very literally.

Practical techniques