Browser Control is AIUNIFY PROS SUITE's autonomous web-research and browser-navigation workspace.
The current interface identifies the feature as:
Browser Control
with:
“Autonomous browser agent interface”.
Browser Control combines several capabilities inside one workspace:
Browser Control is therefore substantially different from AI Chat and the Standard AI Generator.
Instead of simply answering a Prompt from Model knowledge, its source-defined research workflow can:
Search → Navigate → Read → Analyze → Continue Research → Synthesize Findings
within a bounded autonomous loop.
Navigate to:
Sidebar → Core Tools → Browser Control
The current Sidebar route is:
/browser-control
and Browser Control carries both:
New
and:
Pro
interface metadata.
As explained earlier in this manual, the Pro badge itself should not be treated as the sole determination of entitlement. Actual availability depends on your current PROS SUITE account access.
The current feature catalogue classifies:
Browser Control
under:
AI Agents
with a source-defined fallback Token cost of:
25 Tokens.
However, the Browser Control workspace itself does not display a simple:
25 tokens per use
badge.
Its autonomous mission can also make multiple AI requests during one research sequence.
For that reason, this manual does not claim that an entire Browser Control mission will always cost exactly 25 Tokens.
More on Token behavior appears later in this chapter.
Browser Control is best suited for tasks such as:
Primarily:
Ask → AI Responds → Follow Up
Its current built-in utility list is limited to specific Chat tools.
Primarily:
Define Research Objective
↓
Search Web
↓
Navigate to Pages
↓
Extract Page Information
↓
Continue Research
↓
Generate Final Synthesis
Use Browser Control when the task specifically benefits from webpage research rather than ordinary conversation.
PROS SUITE also contains a separate:
Research Agent
under AI Agents.
Browser Control should not be treated as simply another Name for Research Agent.
Browser Control has its own visible:
Research Agent is a separate PROS SUITE Tool and will be documented in its applicable chapter.
The current Browser Control interface is divided into two primary areas.
Research controls and activity:
Browser environment:
The upper-left header displays:
Browser Control
and:
Autonomous browser agent interface.
Below it is the main research-control area.
The primary Prompt area is labeled:
Research Objective.
This is where you explain what you want Browser Control to investigate.
The current Prompt field displays:
“Enter natural language research task... (e.g. 'Compare pricing of Apple Vision Pro on Best Buy vs Amazon')”
You do not need to write browser commands or code.
Describe the research outcome in normal language.
A useful request might be:
Research the pricing, core features, target customer, and enterprise positioning of three leading AI CRM platforms. Use current public sources and give me a comparison with key advantages and disadvantages.
This gives the Agent:
A request such as:
Research CRM.
is extremely broad.
The Agent may need to decide:
A more complete objective usually produces a more focused research process.
For deeper research, include:
What are you researching?
Which products, companies, documents, or markets?
What specifically should be discovered?
For example:
For example:
What should the final answer look like?
Research Company X using its public website. Identify its products, target market, pricing information, leadership positioning, main differentiators, and any publicly stated enterprise capabilities. Summarize the findings for a B2B competitive review.
Compare the current public pricing and major features of Product A, Product B, and Product C. Prioritize each company's official pricing and product pages, then give me a table-style comparison and recommendation by business size.
Review the official documentation for Platform X and identify how its authentication, API access, rate limits, and webhook functionality work. Summarize the most important implementation considerations.
Inside the Research Objective field:
Enter
without Shift launches the Autonomous Agent.
Because the launch handler responds to Enter only when Shift is not held:
Shift + Enter
can be used to add another line without starting the mission.
This is useful for structured research instructions.
When either a Prompt or reference image exists, Browser Control displays:
Clear all.
Selecting it removes:
It does not run the research Agent.
Browser Control contains a Paperclip control titled:
Upload reference image.
The current file picker accepts:
image/*
meaning the verified attachment type is an image.
The current Browser Control upload logic allows images up to:
5 MB.
If the selected image exceeds 5 MB, Browser Control adds an Error log:
“Image upload failed: File size exceeds the 5MB limit.”
Select a smaller image and retry.
When a valid image is loaded, Browser Control records:
“Image uploaded successfully: [filename] ([size] KB)”
After a reference image has been loaded, the Prompt area displays:
IMAGE READY.
A thumbnail is also displayed beneath the Prompt.
Hover over the image Preview and select the X control.
Browser Control removes the asset and adds:
“Reference image removed from workspace.”
Although Browser Control supports a reference image, the Launch Autonomous Agent button requires non-empty Research Objective text.
Therefore:
An image alone is not sufficient to launch the current Browser Control Agent.
Include a written task explaining what the Agent should research or do with the image.
The reference image is supplied to the AI during the first Agent reasoning request.
It is also included in the first history context when the Agent begins performing Search or Navigate operations.
This lets the initial research objective include visual context.
You could attach a product screenshot and enter:
Identify the product shown in this screenshot, then research the manufacturer's current public specifications, positioning, and pricing. Prioritize the official product website.
Browser Control displays:
Example Prompts
beneath the Research Objective.
The current source defines four examples.
Example:
Apple Vision Pro Specs
The underlying Prompt asks Browser Control to find:
from the official Apple website.
Example:
Next.js 15 Features
The Prompt asks the Agent to search official Next.js blog/documentation information and summarize major features and changes.
Example:
Compare Retail Prices
The Prompt asks the Agent to compare iPhone pricing across Best Buy and Amazon.
Example:
GitHub Trending Today
The Prompt asks Browser Control to navigate to GitHub Trending and summarize the top three trending repositories.
Selecting an Example Prompt places its research instruction into the:
Research Objective
field.
It does not automatically start the Autonomous Agent.
Review or customize the Prompt first, then launch it.
Each source-defined preset also contains a startUrl, but the current Example Prompt button only places the preset's Prompt into the Research Objective.
The click handler does not navigate to that preset's stored start URL.
Therefore:
Selecting Apple Vision Pro Specs, for example, does not itself navigate the visible Browser to Apple.com.
The Agent can still search/navigate there as part of its research.
When your Research Objective is ready, select:
Launch Autonomous Agent.
This begins the source-defined autonomous research loop.
The Launch button is disabled when:
During a mission, the button changes to:
Running Sandbox Agent ([iteration]/5).
This shows the current Agent iteration.
The Browser Control source currently sets:
MAX_AGENT_ITERATIONS = 5.
The Agent therefore does not run an unlimited research loop.
An iteration is one AI planning/research turn.
During an iteration, the Agent can:
A single iteration can potentially contain one or more returned research tool calls.
While the Agent is active, Browser Control displays:
Loop Active
and:
Iteration [number] of 5.
A progress bar also advances according to the current iteration.
The current research-agent instructions tell the AI that it has two browser actions:
and instruct it to:
When the Agent decides it needs a web search, it performs a:
browser_search
operation.
The interface records:
Querying web: “[query]”.
The current search processor creates result entries containing:
These results are then returned to the AI so it can decide what to investigate next.
If no results are returned, the Agent receives:
“No search results found.”
It can then determine whether another research path is appropriate.
If the search service fails, the research result contains:
Search error: [message].
When the Agent decides to read a particular webpage, it can use:
browser_navigate.
The activity log records:
Routing connection proxy to: [URL].
The Browser navigation process can retrieve:
When page content is returned to the research Agent, the current processor includes up to the first:
10 links
in the formatted page-content result.
The visible Parsed Links tab can display the broader link collection returned for the page.
If Browser Control cannot fetch a page, the research process returns:
Navigate error: [message].
The intended research sequence is generally:
Search
↓
Review search results
↓
Navigate
↓
Read page text
↓
Review page links
↓
Continue investigating if needed
↓
Synthesize answer
This is the defining workflow of Browser Control.
When the research Agent starts, its instructions include:
Current page URL: [current URL].
Therefore, manually navigating to a relevant starting page before launching can help give the Agent additional context.
If you already know the appropriate official website:
This can make the intended source clearer.
Providing a starting page does not force the research loop to remain on that website.
The Agent's instructions permit searching and navigating multiple times.
If your research must stay within one source/domain, say so explicitly in the Research Objective.
Use only the official company website and its linked documentation. Do not use third-party sources.
This gives the Agent a clearer research boundary.
Use each vendor's official pricing and product pages first, then use reputable third-party sources only where a direct comparison or independent review is useful.
Five iterations are a maximum, not a required minimum.
If the AI returns a final text response without additional Browser tool calls, Browser Control records:
“Agent successfully finalized synthesis and delivered findings.”
and stops the loop.
If all five iterations are used and no final answer has yet been produced, Browser Control records:
“Maximum loop threshold met — Generating comprehensive review report...”
After reaching the five-step threshold without an answer, Browser Control makes a final AI synthesis request instructing the Model:
“Based on all the research you've done, please provide your final comprehensive answer now. Do NOT call any tools.”
This final synthesis is separate from the five visible research iterations.
Because the research loop can use five AI iterations and then invoke a separate final synthesis request, one full mission can involve as many as:
five loop requests + one final synthesis request
under the current client workflow.
This matters when understanding Token usage and mission complexity.
The current implementation does not allow Browser Control to research indefinitely.
The five-iteration cap creates a fixed stopping boundary.
For very large research tasks, divide the work into focused missions.
Instead of:
Research everything about the entire global insurance technology market.
consider separate missions such as:
This helps each mission use its limited research loop effectively.
While the Agent is running, the Launch button changes to Running Sandbox Agent, but the current Browser Control source does not establish a user-facing:
Stop
or:
Cancel Agent
control.
Allow the current mission to complete.
Because there is no verified Stop button:
Prepare Objective → Review → Launch
is preferable to launching first and attempting to correct the mission afterward.
Browser Control contains a research activity console titled:
Telemetry Monitor.
This provides visibility into the Agent's research process.
When Browser Control first loads, the logs begin with:
“Browser agent initialized. Ready for instructions.”
When a new autonomous mission begins, the current logs are reset and replaced by:
Agent started — “[Research Objective]”.
Therefore, preserve old logs before launching another mission if you need them.
The current console recognizes six activity types:
A Search entry represents web-search activity.
Example:
Querying web: “...”
A Navigate entry represents movement to a specific webpage through the Browser proxy.
Example:
Routing connection proxy to: ...
An Agent entry represents agent-level mission activity.
The first entry typically identifies the Research Objective with which the mission was launched.
A Result entry identifies successful research completion or final synthesis.
Examples include:
Agent successfully finalized synthesis and delivered findings.
and:
Final research data synthesized.
An Error entry indicates a research/process problem.
One current example is:
“Model returned empty payload response. Terminal process aborted.”
System logs provide progress and status information such as:
Step 2/5 — Analyzing current viewport & planning next action...
or information related to returned search/navigation results.
Each console entry includes a timestamp with:
based on the user's current browser environment.
As new entries are added, Browser Control automatically scrolls toward the newest log.
This makes it easier to follow active research.
While the Agent or underlying AI system is working, the console can display:
“Awaiting remote LLM instruction packet...”
This is a waiting-state indicator.
The Telemetry Monitor contains a Copy control titled:
Copy Console Output.
Use it when you want to preserve the current research activity logs.
The copied text formats each log approximately as:
[time] [TYPE] content
and combines the entries line-by-line.
After selecting Copy Console Output, the Copy icon temporarily changes to a Check icon for approximately two seconds.
The current source does not establish a separate toast notification for this action.
Beside Copy is a Trash control titled:
Clear Console.
Clear Console removes the current log collection and restores the single default entry:
“Browser agent initialized. Ready for instructions.”
The handleClearLogs action changes only the Log collection.
It does not clear:
according to the current source.
Clears:
Clears:
Do not confuse these two actions.
The bottom of Telemetry Monitor displays:
LOGS: [number].
This number reflects the current number of entries in the local console.
The console status can display:
STREAMING
when the shared AI stream is active.
Otherwise it displays:
SYS_IDLE.
The console currently displays:
PROXY: SECURE.
This is a user-interface status label associated with the Browser Control proxy workflow.
The current Telemetry bar displays:
TEL_PING: 47ms.
The value 47ms is hard-coded in the current interface rather than dynamically calculated from actual network latency.
Therefore:
Do not use TEL_PING: 47ms as a network-performance diagnostic measurement.
Once Browser Control has produced a final research answer, a lower-left panel appears titled:
Synthesis Results.
This is the main final-answer area.
The final Agent answer is rendered using PROS SUITE's Markdown Renderer.
Depending on the AI response, it can contain:
The Synthesis Results header contains a Copy control titled:
Copy markdown content.
This is the equivalent of Copy Answer for the generated research result.
The Copy action writes the complete current:
agentAnswer
to the clipboard.
It copies the underlying Markdown content.
Copying the Synthesis Results:
It simply copies the existing result.
After copying, the icon changes temporarily from Copy to a Check.
Beside the Copy control is:
Download report (.md).
This creates a Markdown research report from the completed Browser Control mission.
The downloaded report begins with:
# AI Browser Research Report
and includes:
The downloaded Markdown includes:
Research Objective: [your objective]
This helps preserve what the Agent was originally asked to investigate.
The report records:
Engine: [selected Model]
or:
Default
when no selected Model value is available.
The Markdown report records a Timestamp generated with the browser's local date/time formatting.
The current report filename follows:
research_report_YYYY-MM-DD.md.
There is a subtle current-version distinction.
The report's internal Timestamp uses the browser's local time, but the filename's date is generated from an ISO timestamp and therefore uses the UTC date.
Around evening hours in North American time zones, the filename date can therefore be one calendar day ahead of the locally displayed report Timestamp.
This is expected from the current implementation.
The .md report contains:
It does not include the Telemetry Monitor's complete Log collection in the current download implementation.
If you need the logs:
Copy Console Output separately.
For a research/audit workflow where you need both:
The current Browser Control source maintains:
as current-page state.
It does not establish a persistent:
system.
Because no persistent Browser Control mission-storage workflow is established in the supplied source, do not assume the current:
can be recovered after the Browser Control page is recreated.
Preserve important research before leaving.
The current Browser Control workspace does not provide:
Save Research
or:
Save Project.
Use:
to preserve results.
Because there is no source-verified Saved Research Project system, there is also no corresponding:
Delete Research Project
workflow.
Clear Console and Clear All only modify the current workspace state.
Browser Control contains a browser-style environment on the right side.
It begins with the current URL:
in the source-defined initial state.
When the Autonomous Agent is not running, the status badge displays:
Manual mode.
You can navigate manually before or after a research mission.
While the Agent is running, the status changes to:
Agent Pilot Mode.
This indicates that the autonomous research loop is currently active.
The Browser address field displays:
Enter URL to navigate manually...
Enter a webpage address and press Enter.
If you enter an address without starting it with http, Browser Control automatically prefixes:
https://.
For example:
example.com
becomes:
https://example.com
Manual URL navigation does more than update the visible Browser.
The current process also attempts to retrieve:
for use by the Text Reader and Parsed Links tabs.
The Browser toolbar includes a Back arrow.
It becomes available when earlier entries exist in the current Browser Control navigation history.
The Forward arrow becomes available when a later history entry exists.
The Browser toolbar includes a Refresh control.
It reloads the current Browser Control URL through the manual navigation workflow.
While a page is loading, the Refresh icon animates and the address field displays a loading spinner.
During page loading, Browser Control can display:
Proxying Document
with the current URL.
The current page-loading visual state contains a fallback that ends after approximately:
10 seconds
if no Browser-loaded event clears it earlier.
The timeout ends the visual loading state; it should not itself be interpreted as proof that every page loaded successfully.
Hovering over the Lock icon displays:
✓ Connection Secure
and:
“SSL proxy active. Cookies and data routed through isolated API endpoint.”
This is the current Browser Control interface wording.
The supplied user-interface source confirms that the tooltip is displayed and that Browser Control routes pages through its Browser proxy endpoints.
The bundle does not contain enough backend infrastructure detail to independently audit every security property implied by that message.
Therefore, this manual records the interface wording but does not expand it into unsupported technical security guarantees.
The current Browser viewport provides:
These three views are extremely useful for understanding what Browser Control has loaded.
Live viewport displays the proxied webpage visually inside Browser Control.
Use it to inspect the actual page presentation.
While the Agent is active, the Live viewport receives visual boundary/scan-line indicators showing that Browser Control is in Agent Pilot Mode.
These are interface indicators rather than webpage content.
The Browser viewport loads the current page through:
/api/browser/proxy
rather than directly using the original site URL as the iframe source.
Because Browser Control uses a proxied/sandboxed environment, some websites may:
A failed page does not necessarily mean the public website itself is offline.
Select:
Text Reader
to view the page's extracted text content.
This is useful when a webpage contains distracting design elements and you want to focus on the information Browser Control extracted.
Text Reader provides:
Filter text content...
Enter a word or phrase to filter the displayed extracted text.
The current Text Reader filter keeps lines containing your search text using a case-insensitive comparison.
This is a local filter of already extracted page text.
It does not launch another web search.
When Text Reader filtering is active, select:
Reset Filter.
This returns to the complete currently extracted page text.
If nothing matches the Reader filter, Browser Control displays:
“No extracted text matches your search.”
The Text Reader displays the textContent extracted by the Browser proxy.
It should not be interpreted as:
It is a reader representation of extracted page text.
Select:
Parsed Links ([number])
to review the links extracted from the loaded page.
Each Parsed Links card can display:
The Parsed Links tab provides:
Search links or anchor text...
The filter searches both:
case-insensitively.
When a Parsed Links search is active, select:
Clear Search.
Each link has a navigation control titled:
Navigate to URL.
Selecting it:
If no links match the current filter, Browser Control displays:
“No parsed links match your search filter.”
These tabs represent the most recently extracted page.
If you navigate elsewhere:
They are not a persistent multi-page research archive.
Browser Control maintains a current-session list of visited URLs and an active history position.
This supports the Back and Forward buttons.
Navigation history should not be confused with a persistent Browser History page.
The source does not establish restoration of this Browser Control navigation history after the workspace is recreated.
When the Agent performs browser_navigate, it calls the same Browser navigation workflow used by the visible Browser viewport.
Therefore:
You can watch the Browser move to pages the research Agent chooses to inspect.
When the Agent performs browser_search, Browser Control queries its search endpoint and gives those results directly to the Agent.
The visible Browser does not need to display a conventional search-engine results page for every autonomous search.
By contrast, browser_navigate updates the visible Browser destination.
This allows you to see the page being investigated.
The current source does not explicitly disable every manual Browser navigation control while Agent Pilot Mode is active.
However, changing pages manually during an autonomous research mission can make the visible viewport diverge from what you expect the Agent to be analyzing.
Best practice:
Allow the Agent to complete its research before performing unrelated manual navigation.
The current Agent instructions establish only two autonomous research actions:
The current source does not establish autonomous Agent functions for:
Do not assume Browser Control performs those actions merely because it displays a Browser viewport.
The embedded Browser sandbox permits certain webpage scripting/form behaviors, but that does not create an autonomous AI form-submission tool.
The source-verified Agent capability remains Search + Navigate/Read.
Browser Control can research public pricing.
It should not be treated as a purchasing Agent.
For example:
Compare prices
is source-aligned.
Buy the cheapest product for me
is not established by the current Browser Control toolset.
The autonomous tool definitions do not establish:
Use Browser Control primarily for accessible public-page research.
Browser Control cannot be assumed to:
Those functions require dedicated systems.
Browser Control reads the current shared:
selectedModel
from your PROS SUITE account.
The shared AI streaming workflow then sends the current selected Model with each applicable Agent AI request.
If you want a particular available Model for Browser Control:
The currently selected Model is written into the downloaded Markdown report as:
Engine.
This is useful when preserving research performed with different Models.
Because Browser Control uses the shared PROS SUITE AI streaming system, user-configured provider API Keys can be included with the applicable AI request.
If a selected Model requires a missing provider credential, follow Chapters 10 and 11.
The current PROS SUITE feature catalogue defines:
Browser Control — 25 Tokens.
However, this is a feature-level default/fallback definition.
Unlike the Standard Generator, the Browser Control page itself does not show:
25 tokens per use
or:
Your balance: ...
beside the Research Objective.
Therefore, users should monitor their overall Monthly Tokens through the normal PROS SUITE account interface.
The Agent invokes the shared AI stream repeatedly as it plans its research.
The loop can make up to five reasoning requests and, when necessary, an additional final synthesis request.
The Browser Control client does not contain its own explicit /api/tokens/deduct call.
Instead, each AI request is sent through /api/ai/stream with:
tool: "browser-control".
The server-side AI endpoint responsible for final Token accounting was intentionally excluded from the sanitized User Operations source bundle.
Therefore, this manual does not assert that:
Those deductions are not source-verifiable from the supplied user-side bundle.
The shared AI streaming workflow refreshes the authenticated user state after each AI stream completes.
This allows updated Token information to be reflected elsewhere in PROS SUITE.
Because a mission can require multiple AI steps:
The current client does not establish a user-visible individual Token charge for:
Do not assign made-up Token prices to these operations.
Even though Browser Control can research actual webpages, the final Synthesis is generated by an AI Model.
Therefore, the final response can still:
Review important findings.
The Agent receives search-result URLs and webpage content as part of its research history.
However, the current Synthesis Results interface does not establish a separate mandatory:
The final AI response may include links when generated, but this is not guaranteed by the current interface.
For high-value business research:
For price comparisons:
Price information can change rapidly.
Browser Control can help locate and summarize information, but legal/regulatory conclusions should be verified against:
For financial or investment decisions, confirm:
before acting.
Health-related conclusions require appropriate authoritative/professional verification.
The presence of web research does not make the final AI synthesis medical advice.
For high-confidence research, include source instructions such as:
Prioritize the manufacturer's official documentation.
or:
Use the government agency's official website.
or:
Use the company's official investor relations materials.
Browser Control's Agent is instructed to select relevant pages, but relevance is still AI-determined.
For important research, manually inspect the key source.
Check whether:
Check:
The current attachment control is disabled while the Agent runs.
If the Telemetry Monitor reports:
Image upload failed: File size exceeds the 5MB limit.
reduce the file size and retry.
This is expected.
The current research loop has:
MAX_AGENT_ITERATIONS = 5.
If no answer has already been produced, Browser Control then attempts its final comprehensive synthesis.
The current interface can log:
“Model returned empty payload response. Terminal process aborted.”
Possible next steps:
If the Agent throws an exception, Browser Control adds:
Sandbox process failure: [message].
Review the Error log and retry if appropriate.
The Agent can receive:
No search results found.
Try:
If the search service itself fails, the returned research content reports:
Search error: ...
Retry later or start from a direct relevant URL.
Browser navigation can return:
Navigate error: Page fetch failed
or another Navigate error.
Possible causes include:
Remember that Live viewport displays a proxied/sandboxed version of the website.
Some pages may behave differently than they do in your normal browser.
Possible explanations include:
First select:
Reset Filter
if filtering is active.
The Parsed Links tab is disabled when:
Use:
Search links or anchor text...
The filter searches both link text and URL.
The UI automatically clears its loading overlay after approximately ten seconds if the expected loaded event does not arrive.
If the underlying page remains unavailable:
Synthesis Results appears only when agentAnswer contains a completed answer.
Check the Telemetry Monitor for:
Copy Answer requires a completed Synthesis result.
If no agentAnswer exists, the Copy handler returns without performing an action.
Report Download also requires a completed agentAnswer.
Wait for Synthesis Results to appear first.
This is expected.
The report's MIME type is:
text/markdown
and its filename ends in:
.md.
The current filename uses the UTC ISO date, while the report Timestamp uses local browser time.
Around local evening hours, those dates can differ.
This is current-version behavior.
This is also expected.
The current Markdown report contains the Synthesis Answer and mission metadata, not Telemetry logs.
Use:
Copy Console Output
when you need the research activity record.
Clear Console intentionally resets the logs to the default initialization entry.
There is no source-verified Undo for this action.
Copy important logs before clearing.
The current mission-start handler replaces the existing Log collection with the new:
Agent started — ...
entry.
Preserve old logs before launching another research mission.
The current Browser Control source does not establish persistent mission storage.
Download important reports before leaving the workspace.
A separate citation system is not source-verified for Synthesis Results.
If source traceability matters:
For example:
Include the URL for every major source used in the final report.
Because automatic citations are not guaranteed, include:
At the end, list the important source URLs used for the research.
This gives the AI a clear output requirement.
When the source is known:
This can reduce unnecessary searching.
Example:
Prioritize official sources. Use third-party sources only for independent comparison.
This improves the research strategy.
For time-sensitive research, write:
Find the currently published pricing as of today.
or:
Use the latest official release information available on the vendor's website.
Then verify the date in the underlying sources.
Pricing and product availability can differ by country.
For example:
Compare current U.S. pricing in USD.
is more precise than:
Compare pricing.
Use:
iPhone 16 Pro Max 256GB
rather than:
iPhone 16
if exact pricing is important.
For easier reuse, ask for:
Browser Control's Markdown Synthesis can format this structure cleanly.
Use all three right-side views:
Live viewport
→ What the actual webpage looks like
Text Reader
→ What text was extracted
Parsed Links
→ What links were found
This is more reliable than relying solely on the AI summary.
After an important mission:
Do this before starting another mission.
Because Browser Control has a five-iteration research limit, a narrow objective usually performs better than an enormous multi-industry assignment.
Break complex work into logical research modules.
If the first report is strong but missing one item, consider launching a narrower follow-up mission rather than repeating the entire research assignment from scratch.
Browser Control can perform multiple AI reasoning steps during one mission.
Review Monthly Tokens before and after larger research tasks.
Complex research may benefit from a stronger reasoning Model, but Model choice can also affect provider access and usage.
Select the desired Model before launching.
Because Download Report automatically records the selected:
Engine
the .md report can help you identify which Model generated a particular research synthesis.
Telemetry Monitor is useful for observing:
activity.
However, it should not be interpreted as an independently certified forensic audit trail.
Some interface labels and values—such as TEL_PING: 47ms—are static presentation elements.
Browser Control is designed around public web research.
Do not unnecessarily place into the Research Objective:
especially when they are not required for the task.
A complete workflow is:
OPEN BROWSER CONTROL
↓
Optional:
MANUALLY NAVIGATE TO STARTING URL
↓
RESEARCH OBJECTIVE
↓
Optional:
UPLOAD REFERENCE IMAGE
↓
Optional:
SELECT EXAMPLE PROMPT
↓
LAUNCH AUTONOMOUS AGENT
↓
Agent loop:
ANALYZE
↓
SEARCH
↓
NAVIGATE
↓
READ
↓
SEARCH / NAVIGATE AGAIN AS NEEDED
↓
Up to:
5 ITERATIONS
↓
If needed:
FINAL SYNTHESIS
↓
SYNTHESIS RESULTS
↓
Review evidence using:
LIVE VIEWPORT
TEXT READER
PARSED LINKS
↓
Preserve:
COPY ANSWER
DOWNLOAD REPORT (.MD)
COPY CONSOLE OUTPUT
Before launching:
While Browser Control is active:
Before relying on important findings:
Before leaving or starting another mission:
.md report.The source verifies subscriber-facing capabilities for:
The supplied user-facing source does not establish:
These capabilities should not be described as active subscriber functions unless a later user-facing release verifies them.
Browser Control can be thought of as having two operating modes.
You control:
The AI research process can:
The current status badge identifies which mode is active.
The current subscriber experience can be summarized as:
BROWSER CONTROL
Autonomous browser agent interface
↓
RESEARCH OBJECTIVE
Enter natural language research task...
↓
Optional:
REFERENCE IMAGE
↓
LAUNCH AUTONOMOUS AGENT
↓
LOOP ACTIVE
Iteration 1–5
↓
Possible actions:
SEARCH
NAVIGATE
READ
↓
Visible in:
TELEMETRY MONITOR
and:
LIVE VIEWPORT
↓
Research inspection:
TEXT READER
PARSED LINKS
↓
When enough information exists:
SYNTHESIS RESULTS
↓
Preserve:
COPY ANSWER
DOWNLOAD REPORT (.MD)
COPY CONSOLE OUTPUT
AIUNIFY PROS SUITE Browser Control is a specialized autonomous web-research workspace labeled:
Browser Control
“Autonomous browser agent interface.”
Users define a natural-language:
Research Objective
and launch:
Launch Autonomous Agent.
The current Agent architecture is bounded to:
5 research iterations
and is instructed to use two research actions:
browser_searchbrowser_navigateto locate, navigate to, read, and analyze public webpage information.
If the Agent reaches the end of its five-step loop without already producing a final answer, Browser Control performs a separate final synthesis request instructing the Model to deliver a comprehensive answer without additional Browser tools.
The workspace provides a Telemetry Monitor with:
activity types. Logs can be preserved using Copy Console Output or reset using Clear Console.
The right-side Browser supports:
Text Reader can locally filter extracted page text, while Parsed Links can search link Names/URLs and navigate directly to a selected link.
When research is complete, Synthesis Results displays the AI's final Markdown answer. Users can copy the result or download:
research_report_YYYY-MM-DD.md
containing:
The feature catalogue currently defines Browser Control with a fallback cost of 25 Tokens, but Browser Control itself can make multiple AI-stream calls during one mission, and the server-side Token-accounting behavior is outside the sanitized source bundle. Accordingly, this manual does not claim a fixed total Token charge for an entire Browser Control mission.
Most importantly, Browser Control's verified autonomous capability is a research workflow:
Search → Navigate → Read → Analyze → Synthesize
It is not source-verified as a general robotic Browser capable of logging into arbitrary accounts, entering passwords, completing purchases, submitting arbitrary forms, or executing transactions.
The recommended operating principle is:
Define a Focused Objective → Prioritize Reliable Sources → Launch the Agent → Watch the Research Process → Inspect the Underlying Pages → Verify Important Findings → Download the Report and Preserve Source Information