What is Feluda Studio?

What is Feluda Studio?

Feluda Studio is the visual workspace where you build AI workflows.

Instead of writing code, you place blocks on a canvas, configure what each block should do, and connect the blocks in the order the task should run.

A workflow can be simple, such as summarising a piece of text, or more detailed, with several AI steps, decisions, tools, and different possible paths.

Studio helps you turn a task that works in Workbench into a process you can save, test, and use again.

What Studio is for

Use Studio when a task:

• contains several clear steps;
• needs to be repeated;
• should produce a consistent result;
• uses more than one AI action;
• needs to make a choice based on the input;
• uses a tool or Gene capability; or
• should be saved for later use.

Workbench is better for exploring an idea through conversation.

Studio is better when you already understand the task and want to organise it into a repeatable process.

How a visual workflow works

A workflow in Studio is built from blocks and connections.

Each block represents one part of the task.

A connection shows where information should go next.

For example:

Input → Summarise with AI → Output

In this workflow:

• the Input block receives the information;
• the AI block prepares the summary; and
• the Output block returns the result.

The visual layout lets you see the complete process at a glance.

The Studio canvas

The canvas is the main working area in Studio.

You use it to:

• add blocks;
• move blocks into a clear layout;
• connect one step to another;
• open block settings;
• review the order of the process; and
• understand how information moves through the workflow.

The canvas can be rearranged without changing the task itself, as long as the required connections remain correct.

Keep related steps close together and arrange the flow in a direction that is easy to follow.

Workflow blocks

Blocks are the building pieces of a workflow.

Different blocks have different purposes.

Common block purposes include:

Block purpose	What it does
Input	Receives text, files, or other information
AI task	Sends an instruction to a selected AI model
Decision	Chooses which path the workflow should follow
Tool	Uses an available capability or connected service
Transform	Changes or organises information
Output	Returns the final result

The blocks available in your Studio may depend on your Feluda version, installed Genes, and configured connections.

You do not need to use every type of block in one workflow.

Connections between blocks

Connections define the order in which information moves.

A connection can pass the result from one block into the next block.

For example:

Customer message
→ Identify the main issue
→ Prepare a draft reply
→ Return the result

Each step receives information from the step before it.

A missing or incorrect connection can prevent the workflow from producing the expected result.

Review the connections before testing the flow.

Configure a block

Each block has settings that explain what it should do.

Depending on the block, you may need to choose:

• the information it should receive;
• the AI provider and model;
• the instruction for the task;
• the tool it should use;
• the condition it should check; or
• the result it should return.

Use clear names and instructions so the purpose of each block remains easy to understand.

For an AI block, an instruction might be:

Summarise the input in no more than five bullet points.

Include decisions, action items, owners, and deadlines.
If information is missing, write "Not provided."
Do not add details that are not in the input.

Test instructions in Workbench before adding them to a repeated workflow.

Choose a model for each AI step

AI blocks can use an available cloud or local model.

When configuring an AI step, choose a model that fits the task.

For example:

• a fast model for simple classification;
• a model with strong instruction following for structured extraction;
• a local model for work you prefer to process on your computer; or
• a model with reliable tool support when the step needs a tool.

A workflow can use different models in different AI steps.

Begin with one model when possible. Add another only when testing shows a clear benefit.

Add decisions and different paths

Some workflows need to respond differently depending on the input.

A decision block can direct the workflow along different paths.

For example:

Incoming message
→ Classify the message
→ If complaint: prepare an apology
→ If question: prepare an answer
→ If other: request human review

Decisions are useful when the categories are clear and each path has a defined purpose.

Test every possible path before relying on the workflow.

Use tools in Studio

A workflow can use tools that are available through Feluda, installed Genes, or supported connections.

A tool may help the workflow:

• retrieve information;
• read or write a supported file;
• create a Journal entry;
• use an approved data source;
• interact with a connected service; or
• complete another supported action.

Review the tool before adding it to a workflow.

Check what information it may receive and whether it performs a read or write action.

Important write actions should remain easy to review.

Use Gene capabilities

A Gene may add tools, prompts, flows, resources, or settings that can support your work in Studio.

You may be able to:

• add a Gene-provided tool to a workflow;
• open an included flow;
• use a provided prompt as a starting point; or
• build on a reusable capability.

Read the Gene description before using its contents.

Confirm any required settings and test the capability with non-sensitive information first.

Save your workflow

Give the workflow a clear name before saving it.

A useful name describes the outcome, such as:

• Summarise Meeting Notes;
• Classify Customer Messages;
• Prepare Weekly Project Update; or
• Extract Contract Dates.

Avoid names such as "Test 1" or "New Flow" when you expect to keep the workflow.

Add a description when available so you can remember:

• what the workflow does;
• what input it expects;
• what result it returns; and
• any model, tool, or connection it requires.

Test the workflow

A workflow should be tested before regular use.

Begin with a simple example.

Then test:

• normal input;
• short input;
• missing information;
• unexpected information;
• a different format; and
• each decision path.

Check whether every block receives the expected information and produces a useful result.

One successful run does not show that the workflow will handle every example correctly.

Run a workflow from Studio

Studio can be used to test the workflow while you are building it.

Running the flow helps you see whether:

• the blocks are connected correctly;
• the input reaches the right step;
• the AI follows its instruction;
• a decision uses the correct path;
• a tool completes its action; and
• the final output is useful.

Review errors and intermediate output when available.

Correct the first step that produces an unexpected result.

Use RunFlows for regular use

After a workflow has been saved and tested, it can be used through RunFlows.

RunFlows is the area where you select a saved flow, provide its input, and review the result.

A simple journey is:

1. build the workflow in Studio;
2. test it with several examples;
3. save it;
4. open RunFlows;
5. provide new input; and
6. review the final output.

Studio is where you design the process.

RunFlows is where you use the saved process.

A simple Studio example

Imagine that you want to turn meeting notes into a short update.

The workflow could contain:

1. an Input block for the meeting notes;
2. an AI block that creates the summary;
3. an AI block that extracts action items; and
4. an Output block that returns both results.

The flow may look like:

Meeting Notes
→ Create Summary
→ Extract Actions
→ Final Output

Each AI block should have one clear purpose.

Separating the tasks makes the workflow easier to review and improve.

Keep each block focused

Avoid asking one block to perform too many unrelated actions.

Instead of one AI block that must summarise, classify, extract, compare, and write a report, use separate steps when the task becomes difficult to review.

Focused blocks make it easier to:

• understand the process;
• find the cause of an error;
• change one part without affecting everything;
• compare models for a specific task; and
• reuse a successful step later.

A simple workflow is usually easier to maintain than a large workflow with unclear blocks.

Name blocks clearly

Give blocks names that describe their purpose.

Use names such as:

• Receive Notes;
• Identify Decisions;
• Extract Action Items;
• Check for Missing Deadlines; or
• Prepare Final Summary.

Clear names make the canvas easier to understand.

They also help when reviewing workflow activity and errors.

Arrange the canvas clearly

A tidy canvas makes the workflow easier to maintain.

Helpful practices include:

• place the starting block on the left or at the top;
• keep the main path in one direction;
• place decision paths near the decision block;
• avoid crossing connections when possible;
• keep related steps together; and
• leave space for future changes.

The layout should help another reader understand the flow without opening every block.

Handle missing information

Decide what the workflow should do when information is missing.

An AI instruction can say:

If a required detail is missing, write "Not provided."
Do not guess.

A decision step may direct incomplete input to a review path.

For example:

Input
→ Check required details
→ Complete: continue
→ Incomplete: request human review

Planning for missing information makes the workflow more reliable.

Plan for errors

A model, tool, provider, or connection may become unavailable.

Consider what should happen when a step fails.

Depending on the workflow, you may:

• stop the flow;
• return a clear error message;
• send the task to a review path;
• use another available model; or
• ask the user for corrected input.

Do not hide an important failure behind a normal-looking final answer.

Make errors visible so they can be reviewed.

Review sensitive information

Before using confidential information in Studio, review the complete workflow.

Check:

• which providers and models receive the information;
• whether a model is local or cloud-based;
• which tools are available;
• whether any step connects to an outside service;
• where the final result is stored; and
• whether unnecessary details can be removed.

A workflow may contain both local and online steps.

Review every step rather than assuming the entire flow follows the privacy level of one model.

Keep a human in the loop

Some workflows should include a person before the final action.

Human review is especially important when the workflow affects:

• customers;
• employees;
• money;
• contracts;
• legal rights;
• health;
• safety;
• security; or
• access to important services.

A workflow can prepare a draft, summary, recommendation, or structured result without making the final decision automatically.

When a workflow is ready

A workflow is ready for regular use when:

• every block has a clear purpose;
• all required connections are present;
• the expected input is understood;
• several different examples have been tested;
• every decision path has been checked;
• tool actions have been confirmed;
• errors are visible; and
• the final result can be reviewed.

Continue improving the workflow when the task or source information changes.

When to use Studio instead of Workbench

Use Workbench when:

• the task is exploratory;
• you expect follow-up questions;
• the process changes often;
• you are comparing instructions; or
• you only need the result once.

Use Studio when:

• the task has clear steps;
• you expect to repeat it;
• the result needs a consistent format;
• tools or decisions are part of the process; or
• the workflow should be saved for RunFlows.

A common path is to discover the task in Workbench and organise it in Studio.

Start with a small workflow

Your first Studio workflow does not need branches, several models, or many tools.

Begin with:

Input → AI Task → Output

Test it until the result is useful.

Then add one improvement at a time.

This makes Studio easier to learn and keeps every change understandable.