Arti Cookbook
How Arti works, the difference between Chat and Fast mode, example prompts, and tips for getting the best results.
Arti Cookbook
Arti is the AI assistant built into Helse. It can write into your documents, create and manage projects, run tasks with AI agents, and answer questions — all from the chat panel.
How Arti Works
Arti has two modes depending on the complexity of your request. Both modes are context-aware — they know which document is open and which project you are in.
Chat Mode vs Fast Mode
| Chat Mode | Fast Mode | |
|---|---|---|
| Best for | Writing, questions, single-action tasks | Multi-step workflows, creating multiple resources |
| How it runs | Single round: one response, one action | Plans steps first, then executes each one |
| Feedback | Direct response in chat | Live step-by-step progress with status indicators |
| Streaming | No | Yes — you see each step complete in real time |
| Use when | "Write an abstract for this paper" | "Set up a literature review project with 5 chapter documents" |
Which mode am I in?:
The chat input has a toggle at the bottom of the Arti panel. The lightning bolt icon activates Fast Mode. Chat Mode is the default.
Context Awareness
Arti automatically detects your current context before deciding what to do.
Open Document
If a document is open in the editor, Arti will write directly into it — without creating a new one — unless you explicitly ask for a new document.
Example: You have "Ottoman Trade Networks" open and type "Write a methodology section." Arti writes into that document. It does not search for it or create a new one.
Active Project
If you are inside a project, Arti uses that project automatically for any new documents or tasks it creates.
Example: You are in the "Climate Migration Literature Review" project and type "Create a task to summarise the latest IPCC reports." Arti creates the task inside "Climate Migration Literature Review" without asking.
Overriding Context
Context awareness has one important caveat: it will never override an explicit instruction.
| You say | Arti does |
|---|---|
| "Write a conclusion" | Writes into the currently open document |
| "Create a new document about conclusions" | Creates a new document |
| "Add a task" | Adds task to the active project |
| "Add a task to a new project called X" | Creates project X first, then adds the task |
Available Tools
Arti can take real actions inside Helse using built-in tools. Here is the full list:
Document Tools
| Tool | What it does |
|---|---|
write_to_active_document | Write content into the currently open document |
create_document | Create a new document in a project |
update_document | Update a document by its UUID |
search_documents | Search for documents by title or content |
list_documents | List documents in a project or folder |
Folder & Project Tools
| Tool | What it does |
|---|---|
create_folder | Create a folder inside a project |
create_project | Create a new project |
list_projects | List all your projects |
search_projects | Find a project by name |
get_project | Look up a project by its UUID |
Task Tools
| Tool | What it does |
|---|---|
create_task | Create a standard task |
list_tasks | List tasks, optionally filtered by project or status |
update_task_status | Change a task's status |
AI Task Tools
| Tool | What it does |
|---|---|
list_ai_agents | List AI agents you own or have access to |
list_cascades | List AI cascades (multi-step workflows) |
create_and_run_ai_task | Create a task, assign it to an agent or cascade by name, and run it |
run_ai_task | Run an existing task with an agent or cascade |
Example Prompts
Writing into an open document
These work best in Chat Mode when a document is already open.
Write an abstract for this paper based on the content in this document.
Add a methodology section covering participant recruitment, data collection, and analysis approach.
Rewrite the introduction to be more concise and foreground the research gap.
Summarise the key arguments from this document into a bullet list at the top.
Write a literature review on the cognitive effects of bilingualism in early childhood education.
Creating documents and projects
These work in both modes. Fast Mode is better when creating multiple things at once.
Create a project called "Antibiotic Resistance Systematic Review" with 3 documents:
a search strategy overview, an inclusion/exclusion criteria summary, and a findings synthesis.
Create a new document in the current project titled "Chapter 3 — Theoretical Framework".
Make a folder called "Primary Sources" in the current project
and add a blank document called "Annotated Bibliography" inside it.
Managing tasks
Create a task called "Review literature review draft" due next Friday with high priority.
Show me all in-progress tasks in this project.
Mark task [task-id] as completed.
List all overdue tasks across my research projects.
AI agent tasks
These require at least one AI agent or cascade to be set up in the AI Builder.
What AI agents do I have?
Create an AI task called "Systematic Review — Climate Migration Literature" and assign it
to the Research Agent to produce a structured synthesis document.
Run a cascade called "Citation Analyser" on the current project
and generate a summary of the key sources and themes.
Create and run a task using the "Writer" agent:
write a 600-word critical synthesis of current debates on urban displacement and housing policy.
Mixed multi-step requests (Fast Mode)
These are ideal for Fast Mode — Arti will plan and execute each step.
Create a project called "Postcolonial Identity in West African Fiction", add three documents
(Theoretical Framework, Close Reading Notes, Draft Argument),
and create 2 tasks: one to develop each core argument section.
Find my "Dissertation" project, create a folder called "Chapter 4 — Empirical Analysis",
and add a document titled "Interview Coding Notes" inside it.
List all my projects, then create a high-priority task in the
most recently updated one called "Supervisor Feedback Review".
Real-World Examples
The following examples show exactly how Arti handles complex multi-step requests in Fast Mode. Each example includes the prompt used and the resulting task execution.
Example 1 — Creating a project with 13 documents
This prompt creates a full book structure in one go. Arti executes 14 tasks: one to create the project, and one per document.
Create a new project called Aphasia. And 13 documents within the project with following titles:
Chapter 1: Fading Margins
Chapter 2: The Parasite in the Blind Spot
Chapter 3: Waking in the Dark
Chapter 4: Two Shadows in Whitehall
Chapter 5: The Pragmatist
Chapter 6: The Paper Black Hole
Chapter 7: Words That Bleed
Chapter 8: Auditors of the Unknown
Chapter 9: The Closing Margin
Chapter 10: A House of Cards
Chapter 11: Into the Aphasia
Chapter 12: The Burden of Remembering
Chapter 13: Echoes in the Architecture
Arti created the project and each of the 13 chapter documents individually, showing live progress as each step completed.
Example 2 — Creating AI tasks to generate files for a project
This prompt creates a project and then uses the default AI agent (HAssist) to generate each file — one AI task per file type.
Create a project called opencode, within this project create a task for the default AI HAssist,
each task should create files for a web app that features a snake game.
Make sure the title ending explains the type of file (.py, .html, .css)
and create these files within the opencode project.
Arti created the project and then created a separate AI task for each file — resulting in 4 tasks: one for .py, one for .html, and one for .css.
Critical: One Task Per Action for AI-Generated Files
Each file needs its own task — otherwise files will be empty:
When asking Arti to generate file content using an AI agent, you must request a separate task for each file. If multiple files are bundled into a single vague task, Arti will create placeholder tasks with no content inside them.
This is the most common mistake when using Arti for code generation or multi-file workflows.
Why it happens
When Arti receives a task to "create all the files", it creates one AI task — but an AI task can only produce one output. The other files either don't get created, or they're created as empty shells waiting for content that never arrives.
Wrong approach
Create a project called Dissertation and generate all the chapter documents
This is ambiguous. Arti doesn't know how many chapters, what they cover, or what each one should contain. The result is typically one task (or empty documents) with no content generated.
Correct approach
Create a project called Dissertation. Create a separate AI task for each of the following documents:
- Task 1: Generate Chapter 1 — Introduction and Research Questions
- Task 2: Generate Chapter 2 — Literature Review on Cognitive Load Theory
- Task 3: Generate Chapter 3 — Methodology and Research Design
- Task 4: Generate Chapter 4 — Analysis and Findings
- Task 5: Generate Chapter 5 — Discussion and Conclusion
Create all documents within the Dissertation project.
Each task is explicit: one AI task maps to one document with a clear scope. Arti can execute each independently and place the output in the right project.
The rule
| You want | How to ask |
|---|---|
| 1 document generated | 1 AI task with a specific title describing the document's content and scope |
| Multiple chapters generated | One AI task per chapter, each with a descriptive title |
| A full research project scaffolded | List every document explicitly — one bullet per task |
Tips for Better Prompts
1. Be specific about where
Vague prompts make Arti guess. Tell it exactly where things should go.
| Instead of | Try |
|---|---|
| "Create a document" | "Create a document called 'Interview Transcripts — Phase 1' in the current project" |
| "Add a task" | "Add a task called 'Revise bibliography' to the Systematic Review project with high priority" |
| "Write something" | "Write a 3-paragraph introduction foregrounding the research gap in studies of multilingual cognition" |
2. Use "current" or "this" to refer to open context
Arti understands:
- "this document" / "the open document" → the document currently in the editor
- "this project" / "the current project" → the project you are browsing
- "here" → generally the most specific active context
Add a conclusion to this document.
Summarise the key arguments in the current document.
Create a task in this project for each chapter in the open document.
3. Separate multiple instructions clearly
Arti can handle several actions in one request — especially in Fast Mode — but clarity helps.
Do three things:
1. Create a folder called "Working Drafts" in the current project
2. Move the open document into it
3. Create a new blank document called "Submission Draft" in the same folder
4. Name agents and cascades as you know them
You do not need UUIDs. Arti looks up agents and cascades by name automatically.
Run the "Research Agent" on a task to write a policy brief on urban displacement.
Use the "Summariser" cascade to create a thematic digest of this project's documents.
5. Use Fast Mode for anything with more than 2 steps
Chat Mode handles one action well. For workflows involving project creation, multiple documents, folder setup, or task batches — switch to Fast Mode for step-by-step execution with live feedback.
Common Mistakes to Avoid
Don't ask Arti to search then update:
Avoid phrasing like "find the Ottoman Trade Networks document and update it." This causes Arti to search and then try to use a placeholder ID. Instead, just open the document first and say "write into this document."
Don't ask for multiple new projects when one is already open:
If you are in a project, Arti will use it. Only say "create a new project" if you genuinely want a fresh one.
AI tasks require an agent or cascade to exist:
Before asking Arti to run an AI task, make sure you have at least one agent or cascade configured in the AI Builder. You can ask "What agents do I have?" to check.
How Fast Mode Works Internally
When you send a message in Fast Mode, Arti runs in two phases:
Phase 1 — Planning
Arti analyses your request and the current context (open document, active project) and produces a step-by-step plan as structured JSON. Each step names a specific tool and its parameters.
Phase 2 — Execution
Each step runs sequentially. If a step produces a result that a later step needs (like a project ID for a new document), the value is resolved automatically. You see each step complete in real time with a status indicator.
Phase 3 — Summary
Once all steps complete, Arti produces a short natural language summary of what was accomplished.
If any step fails, execution stops and the error is shown. Earlier successful steps are not rolled back — so for example if a project was created before a document step fails, the project will still exist.