The Edit Agent feature in Model HQ lets you change and improve agents after they’ve been created. You can make simple updates or completely redesign how the workflow works.

You can:

• Add or remove steps
• Add or remove files used in the agent process
• Change settings such as models
• Update outputs and formats

This makes it easy to update agents as your needs change, without starting from scratch.

There are two ways to edit an agent:

1. Step-by-step editor
• Shows the workflow as a simple list of steps
• Best for quick changes and precise control

1. Visual Builder
• Shows the workflow as a diagram
• Best for visual understanding of the workflow

Both views work on the same agent, so you can switch between them anytime.

This guide will show you how to:

• Open and use the Edit interface
• Understand key sections like inputs, outputs, and settings
• Use action buttons (like Run, Files, Reports, and Controls)
• Edit workflows using the Visual Builder

Learning how to edit agents is important for keeping them up to date, improving performance, and reusing them for different use cases.

1. Launching the edit interface

Editing an existing agent process is straightforward in Model HQ.

The agent to be edited should be selected from the dropdown, and the edit button can be clicked.

Once clicked, an option to use the Visual Builder for editing the agent will be presented.

The Visual Builder shows your agent as a node-and-connection diagram, making it easy to see how steps are linked and how data flows through the workflow.

When you make changes in the Visual Builder, you’ll still review everything in a step-by-step format before saving, so you can clearly confirm the final sequence.

For editing, the step-based editor can be the easiest and fastest option—it makes it simple to add, remove, or adjust individual steps (by adding or deleting rows) without navigating a visual layout.

The choice between the two editing modes is entirely optional. If detailed information about the Visual Builder is required, the Agent Visual Builder Mode documentation should be consulted.

Both editing approaches will be covered in this document. The step-based editor (without Visual Builder) will be explained first.

2. Editing an agent (without visual builder)

This section explains how agents can be edited using the step-based editor. An agent is defined as a linear sequence of steps where each step connects a service, an instruction, and a context. Execution always happens sequentially from the first step to END.

2.1 Details view

The Details view is collapsible by default and provides a high-level, read-only overview of the agent.

The view includes:

• A flowchart visualizing the complete execution path from Start to End
• A numbered list of steps showing the order of execution
• A short description of what each step performs

This view is primarily used to understand the agent flow without modifying it.

2.2 Agent legend

The Agent Legend appears below the Details view and summarizes all key components used by the agent.

The legend includes:

• User Inputs such as text input or image input
• Contexts including files, tables, parsed documents, and intermediate outputs
• Named Variables created during execution and reused across steps

Named variables must be referenced inside instructions using {{variable_name}}.

This section helps track how data moves through the agent.

2.3 Editing steps

Each row in the Agent Builder represents a single step. Each step can be deleted by clicking on the "-" in the same row. A step or a row will be created directly after the current row by clicking "+" to add a step.

The number next to the context column connotes the numbered order of the agent step. This is helpful when referencing that particular step in the agent process in a future agentic step.

Every step is composed of:

• Service Selects the capability used in that step. There is an extensive list of supported services. The services displayed in Service Name depends on the selection in the services button at the bottom of the Step-Based editor or the side nav of the Visual Editor. User must check the service name in order for that particular service to be displayed as an option.

• Instruction A natural language instruction describing what the service should do. (note: some services do not require any input and will be denoted as such) Instructions may also reference named variables {{variable_name}} (i.e., the result of an earlier step in an agent process).

• Context Defines the input source used by the service.

Steps are always executed from top to bottom in the order shown.

Below is the list of supported services, their expected instruction formats, descriptions, and applicable context sources.

2.4 Example: AC field tech support agent

The example below demonstrates an AC Field Tech Support agent configured with 8 sequential steps.

2.4.1 Step-based execution flow

The agent executes steps from top to bottom. Each step can produce outputs that are reused in later steps using named variables.

Execution order in this example:

• Build structured data from CSV
• Read information from an image
• Extract a specific value
• Query structured data
• Query structured data again using a variable
• Parse a technical document
• Answer a question using retrieved context
• End the process

2.5 Action buttons

The Action Buttons provide quick access to configure inputs, manage files, execute the agent, inspect metadata, review outputs, and control the overall agent lifecycle. Each button opens a focused view related to a specific stage of the agent workflow.

2.5.1 Visual editor

The Visual Editor opens the drag-and-drop canvas used to build and edit the agent workflow.

This interface opens the visual builder, where you can:

• Nodes can be added, connected, and reordered
• Services and inputs can be configured
• Execution flow can be visually inspected

2.5.2 Run

The Run action executes a test run of the agent using the currently configured inputs, files, and workflow.

During execution:

• Nodes are executed sequentially from top to bottom
• Intermediate outputs are generated and stored as named variables
• Any configuration or logic issues surface immediately during the run

This mode is primarily used for validation, debugging, and iteration before publishing or sharing the agent.

2.5.3 Inputs

The Inputs button can be used to configure or update the user inputs required to run the agent.

Supported input types:

• MAIN-INPUT – primary text input
• User-Document – documents in various formats
• User-Image – image files such as PNG or JPG
• User-Table – structured data like CSV or JSON
• User-Source – a collection of multiple files treated as one source
• User-Text – short text snippets
• None – no user input required

Selecting the correct inputs is critical. The user must provide values for all selected input types when running the agent. By default, MAIN-INPUT (text) is enabled.

Input-specific notes:

• MAIN-INPUT (text) Free-form text can be pasted into a text field. The current limit is 5000 characters, roughly two pages of text.

• User-Document Used for larger documents. These must be processed first using the parse_document service before being used by downstream services such as rag_answer, semantic_filter, or create_context.

• User-Table Accepts CSV or JSON files. Tables must first be processed with the build_table service to extract data and store it in a local SQL table. Once built, the table can be queried using query_custom_table.

Typical flow:

• User-Image Image files such as PNG or JPEG. Images must first be processed with the vision service, which converts visual content into text-based output that can be reused by later steps.

• User-Text Optional secondary context provided by the user.

• User-Source Allows users to upload multiple files as a single object. Parsing, table building, and vision processing are handled automatically. For best results, it is recommended to apply semantic_filter and create_context early in the workflow.

2.5.4 Files

The Files section is used to upload, manage, and associate assets with the agent.

Capabilities include:

• Uploading documents, images, tables, datasets, or zipped sources
• Assigning file types such as document, image, dataset, table, or source
• Reusing existing datasets or sources already available in the workspace

Uploaded files become available as contexts that can be selected by agent services such as parse_document, build_table, or vision. Multiple files can be added before saving and exiting.

2.5.5 Meta

The Meta section allows descriptive and presentation-related information for the agent to be defined.

This typically includes:

• A short description of what the agent does
• Optional demo links
• A user-facing image associated with the agent

Meta information does not affect execution but is used for documentation, discovery, and presentation.

2.5.6 Outputs

The Outputs section controls what the agent returns after execution.

Options include:

• Standard or custom output selection
• JSON or text-based output formats
• Selecting specific variables, contexts, or intermediate results to expose

This allows fine-grained control over what consumers of the agent actually receive, rather than returning all internal state by default.

2.5.7 Reports

The Reports section controls how agent execution results are packaged into structured, downloadable documents for different audiences such as business users, engineers, or compliance reviewers.

One or more report types can be generated simultaneously:

• Business – high-level summaries focused on outcomes and insights
• Technical – detailed execution logs and system behavior for debugging
• Compliance – safety, audit, and governance records

Each report type allows selection of which elements should be included:

• report – final results or conclusions
• research_list – sources or references used during execution
• response_list – responses generated by each node or step
• safety_record – redactions, filters, and control actions applied
• usage_history – token usage, runtime, and performance statistics
• inference_history – model calls and inference details
• scratch_pad – shared state and intermediate variables
• journal – chronological execution log
• process_map – visual representation of the workflow
• tool_list – tools and services used
• report_commentary – optional notes or annotations

Output formats:

• Word (.docx) – formatted and presentation-ready
• Text (.txt) – plain logs or lightweight exports

Reports can be regenerated at any time after a run.

2.5.8 Controls

The Controls section applies safety, privacy, and content protections automatically during agent execution. These safeguards help prevent sensitive data exposure, unsafe outputs, and malicious inputs without requiring changes to the workflow itself.

Pattern redaction
Sensitive data is automatically detected and masked before it is processed.

Supported patterns include:

• SSN
• ABA routing numbers
• Email
• Credit card
• Driver’s license
• Passport
• Dates
• IBAN
• PAN
• URLs
• Crypto addresses
• Phone numbers

Detected values are redacted or replaced automatically.

Classifier tests
Automated content checks are applied:

• prompt_injection_detection – blocks malicious or manipulative prompts
• toxic_detection – flags unsafe or harmful language
• language_detection – identifies the language of the input
• bias_detection – detects potentially biased or sensitive content

Exclusion list
Custom words or phrases (comma-separated) can be specified to be blocked or ignored during execution.

2.5.9 Services

The Services section is the service catalog for the agent. It determines which capabilities are available when building workflows. Only enabled services can be added as nodes.

Core services
General building blocks for common tasks such as chat, retrieval, extraction, and logic control.

Examples:

• chat – conversational responses
• rag_answer – retrieval-augmented answers
• vision – image understanding
• ocr – text extraction from images
• web_search – online search retrieval
• extract – structured field extraction
• answer – direct question answering
• prompt_builder – dynamic prompt creation
• agent_report – report generation
• embedded_bot – embedded execution
• condition – branching logic
• boolean – rule evaluation

Classifiers
Lightweight text analysis tools for labeling or scoring content.

• sentiment – positive/negative tone detection
• emotions – emotional classification
• topics – topic categorization
• tags – keyword tagging
• intent – intent recognition
• ratings – scoring or grading
• ner – named entity recognition
• summary – concise text summaries
• category – predefined grouping
• q_gen – question generation

Datasets
Tools for preparing, querying, and analyzing structured data.

• select_keys - selects specified keys from a JSON dictionary
• build_dataset – create datasets from JSON
• ds_command_filter - applies filter commands to a dataset
• ds_column_filter - keep rows where a selected column meets your condition
• ds_quick_stats – generates a statistical report based on selected column
• ds_column_analysis - generates a report based on selected column
• ds_report - generate a report of the dataset and the workflow results based on the agent run
• ds_column_select - returns the selcted column from the dataset
• ds_ask_dataset - use a natural language question to retrieve relevant information from the dataset
• ds_readout - returns the text from a set of rows from the dataset for display
• ds_smart_filter – find rows that match the meaning of your query
• ds_keyword_filter - filter rows based on exact text matches in the selected column
• dataset_plot – visualize data
• ds_statistics – perform deeper statistical analysis and generate insights
• load_dataset – load saved datasets
• create_json – provide a list of agent keys to consolidate into a new JSON dictionary
• ds_stat_analysis - generate statistical analysis of input data csv file

Specialized services
Advanced utilities for targeted or complex workflows.

• build_table - create table from CSV data
• query_custom_table - database look-up in natural language
• json_extractor - converts a text chunk with embedded json into a structured dataset element
• semantic_filter – meaning-based filtering
• text_filter – rule-based text filtering
• document_filter – document-level filtering
• table_filter – structured table filtering
• load_kb - load knowledge base into agent state used in 'ask_kb' calls
• ds_ask_kb - answers knowledge base questions from a dataset input
• transformer – text transformation tasks
• aggregate_context - provide a list of context names to consolidate
• parse_document – convert documents to text files
• create_context – build reusable context blocks from the most relevant passages in a source based on query
• report_commentary – generate commentary of key process results from the agent-state - no input context required
• speech_gen – generate audio output from text
• image_gen – generate images from text
• get_stock_summary - stock ticker look-up (requires internet access)
• speech - transcribe a speech file
• speech_batch - transcribe a collection of speec
• vision_batch - answer question based on a collection of image files
• parse_batch - create source from document batch
• extract_tiny - extracts a key-value pair
• website_scraper – extract web content from allowed websites (note: many websites prevent this)
• extract_table – extract tables from documents based on query

Integrations
Connect the agent to external systems or hosted models.

Examples include storage services, email, and external model providers.

Custom services
Workspace-specific or user-defined services added for specialized use cases.

Below is the list of supported services, their expected instruction formats, descriptions, and applicable context sources.

Service Name	Instruction	Description	Context
chat	What is your question or instruction?	Answers a question or performs instruction	MAIN-INPUT, User-Text, None
rag_batch	Enter question or instruction	Performs RAG over batch of documents	User-Document
rag_answer	Ask question to longer document input	Answers a question based on a longer document input	User-Source, Provide_instruction_or_query
vision	Enter question to image file	Provides answer/description from image	User-Image
ocr_vision	Enter instruction	Performs OCR + vision-based understanding	User-Document, User-Image
ocr	Enter name of new document source	Extracts content from image-based or protected documents	User-Document
agent_report	Enter title for agent report	Prepares report on agent output	-
wikipedia_search	Add Wikipedia Articles as Research Context	Adds Wikipedia articles as research context	None
prompt_builder	Enter prompt instruction	Builds structured prompts	None
embedded_bot	Optional	Pauses execution for user interaction	None
condition	Enter expression	Evaluates logical condition	None
web_search	Add query	Performs web search and returns structured results	None
boolean	Provide yes/no question	Provides yes/no answer with explanation	MAIN-INPUT, User-Text
extract	Enter extraction key	Extracts key-value pair	MAIN-INPUT, User-Text
answer	What is your question?	Answers specific question from passage	MAIN-INPUT, User-Text
sentiment	No instruction required	Analyzes sentiment	MAIN-INPUT, User-Text
emotions	No instruction required	Analyzes emotion	MAIN-INPUT, User-Text
topics	No instruction required	Classifies topic	MAIN-INPUT, User-Text
tags	No instruction required	Generates tags	MAIN-INPUT, User-Text
intent	No instruction required	Classifies intent	MAIN-INPUT, User-Text
ratings	No instruction required	Rates positivity (1–5)	MAIN-INPUT, User-Text
ner	No instruction required	Named entity recognition	MAIN-INPUT, User-Text
xsum	No instruction required	Generates extreme summary	MAIN-INPUT, User-Text
summary	Optional	Summarizes content	MAIN-INPUT, User-Text
category	No instruction required	Classifies category	MAIN-INPUT, User-Text
q_gen	No instruction required	Generates questions	MAIN-INPUT, User-Text
build_dataset	Enter dataset name	Create datasets from JSON	JSON Input
select_keys	Enter keys	Select specified keys from a JSON dictionary	JSON Input
dataset_plot	Enter visualization instruction	Visualize dataset	Dataset
load_dataset	Enter dataset name	Load saved datasets	Dataset
create_json	Enter keys list	Consolidate agent keys into JSON dictionary	Agent-State
ds_command_filter	Enter filter command	Applies filter commands to a dataset	Dataset
ds_column_filter	Enter column condition	Keep rows where a selected column meets condition	Dataset
ds_quick_stats	Select column	Generate statistical report based on column	Dataset
ds_column_analysis	Select column	Generate report based on selected column	Dataset
ds_report	No instruction	Generate dataset + workflow report	Dataset
ds_column_select	Select column	Return selected column from dataset	Dataset
ds_ask_dataset	Enter query	Query dataset using natural language	Dataset
ds_readout	Enter row range	Return text from selected rows	Dataset
ds_smart_filter	Enter query	Semantic dataset filtering	Dataset
ds_keyword_filter	Enter keyword	Exact keyword filtering	Dataset
ds_statistics	No instruction	Perform deeper statistical analysis	Dataset
ds_stat_analysis	No instruction	Statistical analysis of CSV dataset	Dataset
build_table	Enter table name	Create table from CSV data	User-Table
query_custom_table	Enter query	Database lookup in natural language	Table Output
json_extractor	Enter schema	Convert embedded JSON text into structured dataset element	MAIN-INPUT, User-Text
semantic_filter	Enter instruction	Meaning-based filtering	User-Source
text_filter	Enter keyword/topic	Rule-based filtering	User-Source
document_filter	Enter document name	Document-level filtering	User-Source
table_filter	No instruction	Structured table filtering	User-Source
load_kb	Enter KB name	Load knowledge base into agent state	None
ds_ask_kb	Enter query	Answer KB questions from dataset input	Dataset
transformer	Choose input	Text/data transformation tasks	Agent-State
aggregate_context	Enter context names	Consolidate multiple contexts	None
parse_document	Enter name	Convert documents to text	User-Document
create_context	Enter instruction	Build reusable context blocks from relevant passages	User-Source
report_commentary	Optional	Generate commentary from agent state	None
speech_gen	Enter text	Generate audio output from text	None
image_gen	Enter description	Generate images from text	None
get_stock_summary	Enter ticker	Stock lookup	None
speech	Enter input	Transcribe a speech file	Audio Input
speech_batch	Enter instruction	Transcribe collection of speech files	Audio Batch
vision_batch	Enter instruction	Answer questions from multiple images	User-Document
parse_batch	Enter instruction	Create source from document batch	User-Document
extract-tiny	Enter key	Extract key-value pair (lightweight)	MAIN-INPUT, User-Text
website_scraper	Enter URL	Extract web content from allowed websites	None
extract_table	Enter query	Extract tables from documents	User-Document
push_to_s3	Enter path	Upload data to S3	None
pull_from_s3	Enter path	Download data from S3	None
connect_library	Enter library name	Connect to semantic library	None
query_library	Enter query	Query semantic library	Library Context
get_quote	Enter symbol	Retrieve stock quote	None
get_company_financials	Enter company/ticker	Retrieve financial data	None
send_email	Enter email	Send email	Select context
openai_chat	Enter instruction	OpenAI chat completion	Text Source
openai_rag	Enter instruction	OpenAI RAG query	Text Source
openai_rag_batch	Enter instruction	OpenAI batch RAG	Text Source
anthropic_chat	Enter instruction	Anthropic chat completion	Text Source
gemini_chat	Enter instruction	Gemini chat completion	Text Source

Service Name

chat

Instruction

What is your question or instruction?

Description

Answers a question or performs instruction

Context

MAIN-INPUT, User-Text, None

Service Name

rag_batch

Instruction

Enter question or instruction

Description

Performs RAG over batch of documents

Context

User-Document

Service Name

rag_answer

Instruction

Ask question to longer document input

Description

Answers a question based on a longer document input

Context

User-Source, Provide_instruction_or_query

Service Name

vision

Instruction

Enter question to image file

Description

Provides answer/description from image

Context

User-Image

Service Name

ocr_vision

Instruction

Enter instruction

Description

Performs OCR + vision-based understanding

Context

User-Document, User-Image

Service Name

ocr

Instruction

Enter name of new document source

Description

Extracts content from image-based or protected documents

Context

User-Document

Service Name

agent_report

Instruction

Enter title for agent report

Description

Prepares report on agent output

Context

-

Service Name

wikipedia_search

Instruction

Add Wikipedia Articles as Research Context

Description

Adds Wikipedia articles as research context

Context

None

Service Name

prompt_builder

Instruction

Enter prompt instruction

Description

Builds structured prompts

Context

None

Service Name

embedded_bot

Instruction

Optional

Description

Pauses execution for user interaction

Context

None

Service Name

condition

Instruction

Enter expression

Description

Evaluates logical condition

Context

None

Service Name

web_search

Instruction

Add query

Description

Performs web search and returns structured results

Context

None

Service Name

boolean

Instruction

Provide yes/no question

Description

Provides yes/no answer with explanation

Context

MAIN-INPUT, User-Text

Service Name

extract

Instruction

Enter extraction key

Description

Extracts key-value pair

Context

MAIN-INPUT, User-Text

Service Name

answer

Instruction

What is your question?

Description

Answers specific question from passage

Context

MAIN-INPUT, User-Text

Service Name

sentiment

Instruction

No instruction required

Description

Analyzes sentiment

Context

MAIN-INPUT, User-Text

Service Name

emotions

Instruction

No instruction required

Description

Analyzes emotion

Context

MAIN-INPUT, User-Text

Service Name

topics

Instruction

No instruction required

Description

Classifies topic

Context

MAIN-INPUT, User-Text

Service Name

tags

Instruction

No instruction required

Description

Generates tags

Context

MAIN-INPUT, User-Text

Service Name

intent

Instruction

No instruction required

Description

Classifies intent

Context

MAIN-INPUT, User-Text

Service Name

ratings

Instruction

No instruction required

Description

Rates positivity (1–5)

Context

MAIN-INPUT, User-Text

Service Name

ner

Instruction

No instruction required

Description

Named entity recognition

Context

MAIN-INPUT, User-Text

Service Name

xsum

Instruction

No instruction required

Description

Generates extreme summary

Context

MAIN-INPUT, User-Text

Service Name

summary

Instruction

Optional

Description

Summarizes content

Context

MAIN-INPUT, User-Text

Service Name

category

Instruction

No instruction required

Description

Classifies category

Context

MAIN-INPUT, User-Text

Service Name

q_gen

Instruction

No instruction required

Description

Generates questions

Context

MAIN-INPUT, User-Text

Service Name

build_dataset

Instruction

Enter dataset name

Description

Create datasets from JSON

Context

JSON Input

Service Name

select_keys

Instruction

Enter keys

Description

Select specified keys from a JSON dictionary

Context

JSON Input

Service Name

dataset_plot

Instruction

Enter visualization instruction

Description

Visualize dataset

Context

Dataset

Service Name

load_dataset

Instruction

Enter dataset name

Description

Load saved datasets

Context

Dataset

Service Name

create_json

Instruction

Enter keys list

Description

Consolidate agent keys into JSON dictionary

Context

Agent-State

Service Name

ds_command_filter

Instruction

Enter filter command

Description

Applies filter commands to a dataset

Context

Dataset

Service Name

ds_column_filter

Instruction

Enter column condition

Description

Keep rows where a selected column meets condition

Context

Dataset

Service Name

ds_quick_stats

Instruction

Select column

Description

Generate statistical report based on column

Context

Dataset

Service Name

ds_column_analysis

Instruction

Select column

Description

Generate report based on selected column

Context

Dataset

Service Name

ds_report

Instruction

No instruction

Description

Generate dataset + workflow report

Context

Dataset

Service Name

ds_column_select

Instruction

Select column

Description

Return selected column from dataset

Context

Dataset

Service Name

ds_ask_dataset

Instruction

Enter query

Description

Query dataset using natural language

Context

Dataset

Service Name

ds_readout

Instruction

Enter row range

Description

Return text from selected rows

Context

Dataset

Service Name

ds_smart_filter

Instruction

Enter query

Description

Semantic dataset filtering

Context

Dataset

Service Name

ds_keyword_filter

Instruction

Enter keyword

Description

Exact keyword filtering

Context

Dataset

Service Name

ds_statistics

Instruction

No instruction

Description

Perform deeper statistical analysis

Context

Dataset

Service Name

ds_stat_analysis

Instruction

No instruction

Description

Statistical analysis of CSV dataset

Context

Dataset

Service Name

build_table

Instruction

Enter table name

Description

Create table from CSV data

Context

User-Table

Service Name

query_custom_table

Instruction

Enter query

Description

Database lookup in natural language

Context

Table Output

Service Name

json_extractor

Instruction

Enter schema

Description

Convert embedded JSON text into structured dataset element

Context

MAIN-INPUT, User-Text

Service Name

semantic_filter

Instruction

Enter instruction

Description

Meaning-based filtering

Context

User-Source

Service Name

text_filter

Instruction

Enter keyword/topic

Description

Rule-based filtering

Context

User-Source

Service Name

document_filter

Instruction

Enter document name

Description

Document-level filtering

Context

User-Source

Service Name

table_filter

Instruction

No instruction

Description

Structured table filtering

Context

User-Source

Service Name

load_kb

Instruction

Enter KB name

Description

Load knowledge base into agent state

Context

None

Service Name

ds_ask_kb

Instruction

Enter query

Description

Answer KB questions from dataset input

Context

Dataset

Service Name

transformer

Instruction

Choose input

Description

Text/data transformation tasks

Context

Agent-State

Service Name

aggregate_context

Instruction

Enter context names

Description

Consolidate multiple contexts

Context

None

Service Name

parse_document

Instruction

Enter name

Description

Convert documents to text

Context

User-Document

Service Name

create_context

Instruction

Enter instruction

Description

Build reusable context blocks from relevant passages

Context

User-Source

Service Name

report_commentary

Instruction

Optional

Description

Generate commentary from agent state

Context

None

Service Name

speech_gen

Instruction

Enter text

Description

Generate audio output from text

Context

None

Service Name

image_gen

Instruction

Enter description

Description

Generate images from text

Context

None

Service Name

get_stock_summary

Instruction

Enter ticker

Description

Stock lookup

Context

None

Service Name

speech

Instruction

Enter input

Description

Transcribe a speech file

Context

Audio Input

Service Name

speech_batch

Instruction

Enter instruction

Description

Transcribe collection of speech files

Context

Audio Batch

Service Name

vision_batch

Instruction

Enter instruction

Description

Answer questions from multiple images

Context

User-Document

Service Name

parse_batch

Instruction

Enter instruction

Description

Create source from document batch

Context

User-Document

Service Name

extract-tiny

Instruction

Enter key

Description

Extract key-value pair (lightweight)

Context

MAIN-INPUT, User-Text

Service Name

website_scraper

Instruction

Enter URL

Description

Extract web content from allowed websites

Context

None

Service Name

extract_table

Instruction

Enter query

Description

Extract tables from documents

Context

User-Document

Service Name

push_to_s3

Instruction

Enter path

Description

Upload data to S3

Context

None

Service Name

pull_from_s3

Instruction

Enter path

Description

Download data from S3

Context

None

Service Name

connect_library

Instruction

Enter library name

Description

Connect to semantic library

Context

None

Service Name

query_library

Instruction

Enter query

Description

Query semantic library

Context

Library Context

Service Name

get_quote

Instruction

Enter symbol

Description

Retrieve stock quote

Context

None

Service Name

get_company_financials

Instruction

Enter company/ticker

Description

Retrieve financial data

Context

None

Service Name

send_email

Instruction

Enter email

Description

Send email

Context

Select context

Service Name

openai_chat

Instruction

Enter instruction

Description

OpenAI chat completion

Context

Text Source

Service Name

openai_rag

Instruction

Enter instruction

Description

OpenAI RAG query

Context

Text Source

Service Name

openai_rag_batch

Instruction

Enter instruction

Description

OpenAI batch RAG

Context

Text Source

Service Name

anthropic_chat

Instruction

Enter instruction

Description

Anthropic chat completion

Context

Text Source

Service Name

gemini_chat

Instruction

Enter instruction

Description

Gemini chat completion

Context

Text Source

2.5.10 JSON editor

The JSON Editor displays the complete agent configuration in raw JSON format.

This view shows the full structure of the agent, including services, nodes, inputs, contexts, and execution settings, allowing precise inspection and modification of how the workflow is defined.

Common sections include:

• service – service assigned to each node
• node – execution order
• process_inputs – required user inputs
• context_id_list – data passed between steps
• sample_files – example inputs
• configuration metadata

Changes made in this editor directly update the agent configuration.

3. Editing an agent (with visual builder)

The Visual Builder allows agents to be edited using a point-and-click, drag-and-drop interface. This makes it easy to understand, modify, and extend agent logic without writing code.

Note: The number displayed on each node represents the order in which they were placed on the canvas by the user only. The number does not indicate the order in which the nodes will run. The sequence of agent action is determined by the connectors between each node in the process, and not the number on the node.

3.1 Builder overview

The canvas represents the full execution flow of an agent, from input to final output. Each block is a node, and connections define how data moves between steps.

Up to Transformers, all components are fully visual and configurable directly in the builder. The remaining options are covered in detail in the dedicated Visual Builder documentation.

3.2 Left panel (node types)

The left sidebar contains the core building blocks that can be dragged onto the canvas:

• Input Defines how data enters the agent (for example text, image, or file input).

• Node General-purpose processing steps that pass data forward.

• Classifier Routes execution based on intent or classification logic.

• Bot Handles LLM-powered reasoning or responses.

• Condition Adds branching logic based on rules or outputs.

• Transformer Transforms or enriches data before it moves to the next step.

3.3 Canvas controls

On the canvas, the following actions can be performed:

• Nodes can be dragged to reposition them
• Nodes can be connected to define execution flow
• Nodes can be selected to edit their instructions and configuration

3.4 Zoom and utility actions

The bottom-left controls allow the following operations:

• Zoom in
• Zoom out
• Clear the canvas

3.5 Action buttons

Below the utility buttons, quick access to key actions is provided:

• Download Agent Downloads the agent definition as a JSON file. This file can later be uploaded to instantly recreate the agent.

• Run Executes the agent with the current configuration.

• Home Returns to the main dashboard.

Additional information about the Visual Builder can be found in the Agent Visual Builder Mode documentation.

Conclusion

This document provided comprehensive guidance on editing agents in Model HQ using both the step-based editor and the Visual Builder interface. The Edit Agent functionality enables existing workflows to be refined, extended, and optimized by modifying services, adjusting execution order, configuring inputs and outputs, and applying safety controls. The step-based editor offers precise control over linear agent workflows through a structured interface where services, instructions, and contexts can be configured for each execution step. The Visual Builder provides an intuitive drag-and-drop canvas for visualizing complex workflows, making structural changes, and understanding data flow through node-and-wire representations.

Key editing capabilities include configuring input types (text, documents, images, tables, sources), managing files and assets, executing test runs for validation, defining metadata and presentation elements, controlling output formats, generating structured reports for different audiences, and applying automated safety controls such as pattern redaction and content classification. The Services catalog determines which capabilities are available within the agent, ranging from core services like chat and RAG to specialized utilities for web scraping, image generation, and email automation. Advanced features such as the Plan section enable natural language descriptions to be used for workflow generation, while the JSON Editor provides direct access to the underlying agent configuration for precise modifications.

For further assistance or to share feedback, please contact us at support@aibloks.com