AI Assistant

The AI Assistant lets you ask questions about your FieldOps data using natural language.

Instead of manually searching through projects, sites, visits, findings, corrective actions, and performance information, you can ask a question and let the assistant retrieve and analyze the relevant operational data.

AI Assistant

Ask questions about your FieldOps data and get answers about projects, sites, visits, findings, actions, and performance.

What the AI Assistant can do

The AI Assistant can help you:

• Understand project performance
• Find sites requiring attention
• Review visits and their results
• Understand findings and unresolved issues
• Identify overdue corrective actions
• Analyze operational performance
• Explore DHIS2 performance data when DHIS2 is configured
• Combine information from different parts of your FieldOps data
• Ask follow-up questions about an earlier answer

For example, you can ask:

Which sites currently require attention?

Or:

Which corrective actions are overdue?

You can also ask more analytical questions:

Why has Kapsabet Sub-County Hospital been underperforming for the last three months?

The assistant can retrieve the relevant information from FieldOps and use it to provide an explanation.

How it works

The AI Assistant combines a language model with controlled access to FieldOps operational data.

A typical request follows this process:

1. You ask a question in natural language.
2. The assistant interprets the question and determines what information is needed.
3. Relevant FieldOps data is retrieved using the available data tools.
4. The assistant analyzes the retrieved information.
5. A natural-language response is returned to you.

This means the assistant does not need to be given all of your operational data in the question itself.

Your organization's data

The assistant operates within your FieldOps organization context.

It uses the data available to your organization and the tools made available to the AI Assistant. It does not provide unrestricted access to the underlying database.

The information available to the assistant depends on:

• The data in your FieldOps workspace
• Configured integrations
• Available monitoring and visit records
• Findings and corrective actions
• Configured project and performance data

AI Assistant is not an autonomous agent

The current AI Assistant is designed to read, analyze, explain, and summarize operational information.

It does not currently:

• Create records
• Edit records
• Delete records
• Approve or reject visits
• Create corrective actions
• Change project configuration
• Independently perform operational actions

Any changes to your FieldOps records continue to be performed through the normal FieldOps workflows.

Examples

You can ask questions such as:

How many sites are in this project?

Which sites require attention?

Show me the unresolved findings.

Which actions are overdue?

What happened during the latest visit to this site?

How is this project performing?

Why has this site been underperforming?

What are the main operational issues across this project?

The assistant is particularly useful when the answer requires information from more than one area of FieldOps.

Getting started

To start using the AI Assistant:

1. Open the AI Assistant from the FieldOps interface.
2. Enter your question.
3. Submit the question.
4. Review the response.
5. Ask a follow-up question if you need more detail.

You can also use the suggested questions provided by the assistant as starting points.

Using the AI Assistant

The AI Assistant is designed to make it easier to explore and understand your FieldOps data through natural-language questions.

You do not need to know database queries, report structures, or technical terminology.

Open the AI Assistant

Open the AI Assistant from the FieldOps interface.

The assistant provides a conversation area where you can enter questions about your organization's operational data.

The assistant is identified as:

AI Assistant
FieldOps data & analysis

Ask a question

Enter your question in the message field and submit it.

For example:

Which sites require attention?

The assistant will analyze the question, retrieve the relevant information, and return a response.

Questions can be simple or analytical.

Simple questions

How many sites do we have?

How many projects are currently active?

Which actions are overdue?

Analytical questions

Why is this project underperforming?

What are the main issues affecting this site?

What patterns can you see in the recent visits?

The more specific your question, the more focused the response can be.

Use suggested questions

The AI Assistant provides suggested questions to help you get started.

Suggested questions can be useful when you are not sure what to ask or want to quickly explore common operational questions.

Select a suggested question to send it to the assistant.

You can then continue the conversation with a follow-up question.

Ask follow-up questions

You do not need to repeat the entire context of a question when continuing a conversation.

For example:

Which sites require attention?

After reviewing the response, you could ask:

Which one has the most unresolved findings?

Or:

When was its last visit?

Follow-up questions allow you to progressively investigate an issue.

Ask focused questions

For better results, include useful context when it is relevant.

Instead of:

How is performance?

Try:

How has the immunization project performed over the last three months?

Instead of:

What is wrong?

Try:

What are the main unresolved issues affecting this site?

Useful context can include:

• Project name
• Site name
• Time period
• Performance area
• Finding or action status

Explore operational issues

The assistant can help you move from a high-level question to a more detailed investigation.

For example:

1. Ask which sites require attention.
2. Identify a site with repeated problems.
3. Ask about its recent visits.
4. Review its findings.
5. Ask about unresolved corrective actions.
6. Ask what factors appear to be contributing to the problem.

This allows you to investigate operational issues conversationally rather than navigating through multiple screens.

Review AI responses

AI-generated responses should be treated as an analysis of the available FieldOps data, not as a replacement for the underlying records.

When a response is important for operational or management decisions, verify the relevant information in FieldOps.

The quality of an answer depends on the quality and availability of the underlying data.

Expand the assistant

The AI Assistant can be expanded when you need more space for a longer conversation or a detailed response.

Collapse it when you want to return your attention to the main FieldOps interface.

Ask another question

You can continue asking questions throughout the conversation.

There is no requirement to use a specific query format. The assistant is designed to understand normal language.

For example:

Show me overdue actions.

followed by:

Which project do they belong to?

and then:

Which of those projects has the most unresolved findings?

Tips for better questions

For useful results:

• Be specific when you know the project or site.
• Include a time period when performance over time matters.
• Ask one clear question at a time for complex investigations.
• Use follow-up questions to drill into the answer.
• Verify important conclusions against the underlying FieldOps records.

The goal is not to learn how to talk to an AI. The goal is simply to ask FieldOps questions in the way you would ask a colleague.

What You Can Ask

The AI Assistant can answer questions across several areas of your FieldOps operational data.

You can ask for information, summaries, comparisons, explanations, or follow-up analysis.

Projects

Ask questions about your projects and their current state.

Examples:

How many projects do we have?

What projects are currently active?

How is this project performing?

Give me a summary of this project.

Which sites belong to this project?

Sites

Use the assistant to understand the status and operational condition of sites.

Examples:

How many sites do we have?

Which sites require attention?

Which sites are performing below target?

What is happening at Kapsabet Sub-County Hospital?

When was the last visit to this site?

Visits

The assistant can help you explore monitoring visits and their results.

Examples:

Show me the latest visit for this site.

What happened during the last visit?

How many visits have been completed?

What findings were recorded during the visit?

Show me the visit history for this site.

Where visit information is available, the assistant can use it to provide a concise operational summary.

Findings

You can ask questions about findings identified during monitoring.

Examples:

Show me unresolved findings.

Which sites have unresolved findings?

What are the main findings for this project?

Which findings have remained unresolved?

What issues were identified during the latest visit?

Findings can also be considered together with visits, sites, and corrective actions to understand broader operational issues.

Corrective actions

Ask about actions that have been created to address findings.

Examples:

Which corrective actions are overdue?

Show me overdue actions for this project.

Which sites have outstanding actions?

How many actions are still unresolved?

What actions are associated with this finding?

This can help managers identify issues that require follow-up.

Performance

The AI Assistant can help explain operational performance rather than simply displaying individual records.

Examples:

How is this project performing?

Which sites are below target?

Which site is performing worst?

Why has this project been underperforming?

What are the main factors affecting performance?

When sufficient data is available, the assistant can connect performance information with visits, findings, and corrective actions.

DHIS2 performance

When DHIS2 is configured and the relevant data is available, you can also ask questions about DHIS2 performance.

Examples:

How is DHIS2 performance for this project?

Which sites are below target?

Which indicators are performing poorly?

Which sites need performance follow-up?

The availability of DHIS2 information depends on the configuration and synchronization of the relevant DHIS2 connection and project indicators.

Cross-domain questions

Some of the most useful questions combine information from different parts of FieldOps.

For example:

Why has Kapsabet Sub-County Hospital been underperforming for the last three months?

Answering a question like this may require looking at several types of information, including:

• Site performance
• Recent visits
• Findings
• Corrective actions
• Performance indicators

Another example:

Which sites are underperforming and also have overdue corrective actions?

This type of question can help identify operational priorities.

Questions about trends

You can ask the assistant to compare information over time.

Examples:

How has this site's performance changed over the last three months?

Has performance improved since the previous visit?

What changed between the last two visits?

Are the same issues appearing repeatedly?

The answer depends on the historical data available in FieldOps.

Questions for investigation

The assistant can also be used as an investigation starting point.

For example:

Which sites require attention?

After identifying a site, you could continue:

What are the main issues at that site?

Then:

Show me its recent visits.

Then:

Which findings remain unresolved?

And finally:

Which corrective actions are overdue?

This creates a conversational path from identifying an issue → understanding it → investigating it → identifying follow-up needs.

Ask in your own words

You do not need to use exact commands.

For example, these questions can express a similar intent:

Which sites need attention?

Are there any sites we should be concerned about?

Show me the sites that need follow-up.

The assistant is designed to interpret natural language and identify the relevant FieldOps information.

When the assistant cannot answer

The assistant may not be able to answer a question when:

• The required data does not exist in FieldOps.
• The relevant integration has not been configured.
• Data has not yet been synchronized.
• The question requires an action that the assistant does not currently perform.
• The requested information is outside the tools available to the assistant.

When this happens, check the relevant FieldOps records, integration configuration, or ask a narrower question.

AI Providers and Models

FieldOps uses a provider-neutral AI architecture. This allows organizations to connect supported AI providers and select available models for their AI Assistant.

FieldOps is not designed to lock an organization into a single AI provider.

Bring Your Own AI Provider

AI provider access is configured by the organization.

This approach is commonly referred to as Bring Your Own Key (BYOK).

Your organization provides the credentials required to access its selected AI provider. The provider then handles the organization's model usage and billing directly.

FieldOps does not act as an intermediary for AI model billing.

Provider-neutral architecture

The AI Assistant is designed to work independently of a specific model vendor.

The AI layer can support providers such as:

• OpenAI
• Google Gemini
• Anthropic
• Mistral
• DeepSeek
• Qwen
• Meta Llama
• Grok

The providers and models available in a particular FieldOps installation depend on the providers configured and the models enabled by the system.

Models

AI providers offer different models with different capabilities.

FieldOps maintains a model catalogue so that available models can be associated with their providers.

A provider is available for selection when it has active models configured for use.

This allows the available model selection to evolve without changing the overall AI Assistant workflow.

Choosing a model

When multiple models are available, the appropriate choice can depend on the type of work being performed.

For example, some models may be better suited to:

• General operational questions
• Longer analytical responses
• Complex reasoning
• Large amounts of context
• Faster responses
• Lower-cost routine questions

The AI Assistant experience remains the same regardless of the selected provider or model.

Provider credentials

Provider credentials are organization-specific.

Keep provider credentials secure and do not include API keys in:

• AI Assistant questions
• Project names
• Visit notes
• Findings
• Corrective actions
• Other operational data

Provider credentials should only be entered through the appropriate AI provider configuration workflow.

Usage and telemetry

FieldOps records AI usage information for operational monitoring and administration.

Usage information can include information such as:

• Provider
• Model
• Feature
• Request
• Token usage
• Request duration
• Request status
• Error information

This helps administrators understand AI usage and diagnose problems.

Provider billing

Your organization remains responsible for its relationship with the selected AI provider.

AI provider charges are handled directly by the provider according to the organization's provider account and usage.

FieldOps does not add AI provider charges to the organization's FieldOps subscription through the AI Assistant.

Provider availability

If a provider or model is not available for selection, it may not currently have an active model configured.

A provider may also be temporarily unavailable because of:

• Invalid provider credentials
• Provider-side errors
• Account limits
• Exhausted provider credits
• Rate limits
• Provider outages

These conditions are separate from the FieldOps AI Assistant interface itself.

Changing providers

Because the AI Assistant is provider-neutral, organizations can change the configured provider or model without changing how users ask questions.

For example, a user can continue asking:

Which sites require attention?

regardless of whether the configured model is provided by OpenAI, Gemini, Anthropic, DeepSeek, Qwen, or another supported provider.

The underlying provider may change, while the FieldOps workflow remains the same.

AI Assistant Data and Tools

The AI Assistant does not have unrestricted access to the FieldOps database.

Instead, it uses a controlled set of tools to retrieve the information required to answer a question.

This provides a structured boundary between the AI model and FieldOps operational data.

What data can the assistant access?

The available information covers operational areas such as:

• Projects
• Sites
• Visits
• Monitoring responses
• Findings
• Corrective actions
• Performance information
• DHIS2 performance data where configured
• Operational timelines

The exact information available depends on the data in your organization and the integrations configured for your workspace.

Controlled data access

When you ask a question, the assistant determines which available data it needs.

For example:

Which sites require attention?

may require site and performance information.

A question such as:

Why has this site been underperforming?

may require several types of information, such as:

• Site information
• Performance data
• Recent visits
• Findings
• Corrective actions

The assistant can retrieve the relevant information and use it to formulate an answer.

Tools

The AI Assistant uses purpose-built tools for common operational queries.

Examples of available capabilities include retrieving:

• Projects
• Sites
• Project summaries
• Sites belonging to a project
• Sites requiring attention
• DHIS2 performance
• Site operational summaries
• Visit details
• Visit findings
• Visit actions
• Visit timelines
• Overdue actions
• Action searches
• Visit searches

These tools provide structured access to FieldOps information.

Organization context

FieldOps is a multi-organization platform.

AI Assistant requests operate within the user's active organization context.

This is important because the same question can produce different results for different organizations.

For example:

How many sites do we have?

returns information relevant to the organization in which the user is currently working.

The assistant does not query the database directly

The language model is not given unrestricted SQL or direct database access.

Instead, FieldOps controls the operations that can be requested by the assistant.

This allows FieldOps to define:

• What information can be retrieved
• How information is retrieved
• Which operations are available
• What data is returned to the model

The assistant then uses those results to produce its response.

Multiple tools for one question

A single question can require more than one data lookup.

For example:

Why has Kapsabet Sub-County Hospital been underperforming for the last three months?

may require the assistant to combine information from performance, visits, findings, and actions.

The assistant can use the relevant tools and then synthesize the returned information into one response.

Source data remains authoritative

The AI Assistant is an interpretation layer over your operational data.

The underlying FieldOps records remain the source of truth.

For important decisions, use the assistant to identify and understand relevant information, then verify the conclusion against the corresponding FieldOps records.

Data availability matters

The assistant can only analyze information that is available to it.

For example, a question may produce an incomplete answer when:

• Visits have not yet been recorded.
• An integration has not been synchronized.
• A project has limited historical data.
• Findings have not been entered.
• Corrective actions have not been created.
• Required performance data is unavailable.

A missing answer does not necessarily mean that there is no underlying operational issue. It may mean that the required data is not available.

Integrations

External data sources can expand the information available in FieldOps.

Depending on your configuration, this can include data from:

• KoboToolbox
• ODK Central
• DHIS2
• Excel imports

Once supported data has been brought into FieldOps, the relevant information can become part of the operational context available to the AI Assistant.

Read and analyze, not act

The current toolset is designed primarily for retrieval and analysis.

The assistant can help you understand:

What happened?

What is happening?

Where is the problem?

What requires attention?

What patterns are visible?

It does not currently use these tools to independently modify FieldOps records or perform operational actions.

This boundary is intentional: users remain in control of changes to operational data.

AI Assistant Limitations

The AI Assistant is designed to help users understand FieldOps data, but AI-generated responses have limitations.

The assistant should be used as an analytical aid rather than as an unquestionable source of truth.

AI responses can be incorrect

AI models can sometimes produce inaccurate or incomplete responses.

An answer may be incorrect because:

• The model misunderstood the question.
• The available data was incomplete.
• The retrieved information did not contain everything needed.
• The model made an incorrect inference.
• A provider returned an unexpected response.

For important operational decisions, verify the answer against the underlying FieldOps records.

The assistant can only use available data

The AI Assistant cannot reliably answer questions about information that is not available to it.

For example, it may not be able to determine why a site is underperforming if the required performance, visit, finding, or action data has not been recorded.

The assistant does not create missing historical information.

External integrations affect data availability

When your analysis depends on an external system, the relevant integration must be configured and its data must be available in FieldOps.

For example, DHIS2-related questions depend on the relevant DHIS2 connection, synchronization, mappings, and performance data.

Similarly, data collected through KoboToolbox or ODK Central must be available in FieldOps before it can be used as part of the assistant's analysis.

AI provider availability

The AI Assistant depends on the configured AI provider.

A request may fail when the provider experiences:

• An outage
• Rate limiting
• Authentication problems
• Account restrictions
• Insufficient credits
• Usage limits
• Other provider-side errors

These failures do not necessarily indicate a problem with the underlying FieldOps data.

Model differences

Different AI models can produce different responses to the same question.

They may differ in:

• Reasoning ability
• Response length
• Interpretation of ambiguous questions
• Speed
• Handling of complex analysis

For this reason, changing the configured provider or model can change the wording or analysis of an answer even when the underlying FieldOps data remains unchanged.

No autonomous operational actions

The current AI Assistant does not independently modify your FieldOps data.

It does not currently:

• Create records
• Edit records
• Delete records
• Approve visits
• Reject visits
• Create corrective actions
• Close corrective actions
• Change project configuration

Users continue to perform these operations through the normal FieldOps workflows.

Permissions and organization context

The assistant operates within the FieldOps organization context available to the user.

Its responses depend on the information and tools available in that context.

Users should continue to follow their organization's normal access-control and data-governance practices.

Responses are not a replacement for source records

The assistant provides a conversational interpretation of operational information.

The original FieldOps records remain authoritative.

When a response affects:

• Programme decisions
• Management decisions
• Compliance reporting
• Performance reviews
• Corrective actions
• Donor reporting

verify the relevant information against the underlying records before taking action.

Ambiguous questions

Broad questions can have multiple reasonable interpretations.

For example:

How is the project doing?

could refer to:

• Site performance
• Visit completion
• Findings
• Corrective actions
• DHIS2 indicators
• Overall operational status

If the answer is not what you expected, make the question more specific.

For example:

How has site performance changed over the last three months?

or:

Which sites are below target?

Getting better results

You can often improve an answer by providing more context.

Include, where relevant:

• A project
• A site
• A time period
• A performance indicator
• A finding
• An action status

You can also ask follow-up questions to narrow down the analysis.

Treat AI as an analysis layer

The best way to use the AI Assistant is as a conversational layer over FieldOps.

Use it to:

• Explore operational data
• Find issues
• Summarize information
• Investigate patterns
• Identify areas requiring attention
• Ask questions that would otherwise require multiple screens or reports

Then use FieldOps's underlying records and workflows to verify information and take action.