Chat With Your Project Data: How AI Uses Your Context
2 days agoPUBLISHED INAgile
|
Quick answer To chat with your project data, an AI tool needs context: the stories, bugs, requirements and documents it's allowed to read. In Sanplex, you add that context with @mentions, and the data comes from a vectorized copy of your project that refreshes on synchronization. That means answers can be fast and specific, but not live, and permissions need a careful look. Always check an answer against the source item. |
A chat window that knows nothing about your project gives you generic answers. Ask it which bugs block the release and it'll politely explain what a blocker is.
The fix is context. This guide explains how an AI tool uses your project data to answer, how Sanplex handles it with @mentions and data vectorization, and how to check that what comes back is right. We'll say where Sanplex fits, and where its limits are.
What does it mean to chat with your project data?
It means the AI answers from your stories, bugs, tasks and documents instead of from general knowledge. The tool gives the model a slice of your data along with your question, and the model writes an answer based on it.
-
Context in. Your question travels with relevant project data.
-
Answer out. The model replies from that data, not from the whole internet.
-
Selection matters. What you include decides what it can answer.
-
Freshness matters. The answer is only as current as the data it was given.
-
People stay responsible. A fast answer still needs a check.
How Sanplex adds context
The Sanplex Agent guides describe two sources of context. One is the page you're on, such as a document you're reading. The other is vectorized Sanplex data: objects like requirements, bugs and documents that have been converted into a searchable form.
|
Mention |
What it brings into the chat |
Example |
|---|---|---|
|
@Current Document |
The document you're viewing |
Ask for a summary of the spec you have open |
|
@Sanplex |
All vectorized object type data, including requirements, bugs and documents |
Ask what changed across the project |
|
@Sanplex object type |
All data of one type |
@Bug to include all bug data |
|
@Sanplex object with an ID |
One specific object |
@Requirement#123 to include requirement 123 |
The narrower the mention, the sharper the answer. A question about one requirement works better with its ID than with the whole project in the room.
What has to be set up first
Chat with project data depends on a service called ZAI, which the Agent relies on. Once ZAI is connected, an admin turns on data vectorization in the Agent module, and Sanplex converts supported data into vectors. After the first synchronization finishes, you can reference Sanplex data in conversations. The ZAI configuration guide walks through the steps, and the Sanplex Agent chat guide covers how to use the chat.
The limits you should know about
The guides are open about three limits, and they affect how you use the feature.
|
Limit |
What it means |
What to do |
|---|---|---|
|
Not real time |
Vectorized context uses the data from the last synchronization |
Check the source item for anything that changed today |
|
No permission control |
Vectorized data may be visible in chat to users who lack permission in Sanplex |
Don't enable it if your access rules are strict |
|
Loaded at the start |
Context is loaded when the conversation is created, on the first message |
Start a new conversation after major changes |
The permission point deserves the most thought. If a standard user can reach requirements or documents in chat that they couldn't open in the tool itself, that's a governance decision, not a technical footnote.
How to check an answer
|
Check |
What to look for |
Why |
|---|---|---|
|
Open the source |
Click through to the story, bug or document |
The answer is a summary, not the record |
|
Check the date |
Was the item changed after the last synchronization? |
Context can be a few hours or days old |
|
Spot counts |
If it says nine bugs are open, count them |
Numbers are where chat slips |
|
Ask for the reference |
Which item did that come from? |
A good answer can name its basis |
|
Try a trap |
Ask about something that doesn't exist |
A sound tool says it can't find it |
Questions that work well with context
-
One item, deep. "Summarize @Requirement#123 and list open bugs that mention it."
-
One type, scan. "Using @Bug, which ones mention export?"
-
A document, quick. "Summarize this spec in five bullets."
-
Cross checks. "Do any requirements conflict with this document?"
-
Weak spots. Counting, anything very recent and anything nobody recorded.
Security and governance basics
-
Who can see what. Understand how permissions behave before you enable it.
-
What gets vectorized. Know which data types are included.
-
Where it runs. Cloud or your own servers, and where the model sits.
-
Pilot narrowly. Start with one project that has no sensitive data.
-
Records. Note who enabled it and why.
How to measure success
-
Correct answers. Test twenty questions and score them against the source.
-
Time saved. Compare with filters and manual searching.
-
Stale answers. How often a changed item caused a wrong reply.
-
Adoption. Whether people use it weekly.
Build, buy or generate?
|
Approach |
Good for |
Watch out for |
|---|---|---|
|
Chat built into your project tool |
Teams that want answers next to the work |
Limited to the tool's own data, with its limits |
|
Separate AI tool over exports |
Quick experiments |
Stale data and no live link |
|
Custom build with an API |
Teams with engineering time |
Permissions, upkeep and security are on you |
|
General chatbot, paste items |
One off questions |
No context beyond what you paste |
How Sanplex fits
Sanplex is a full lifecycle project management tool and a Jira alternative for Agile, Waterfall and CMMI teams, built by EasySoft INC. Scrum, Kanban, stories, tasks, bugs, test case management, documents, product planning and reports all live in the one tool.
Sanplex Agent chat supports conversations with language models, model switching and @mentions of Sanplex objects as context. For documents in particular, see our piece on an AI assistant for project documentation. Sanplex Agent, the AI assistant, writes story descriptions, creates bug reports from keywords and turns stories into test cases. It's available in the Standard and Premium plans.
On price, Sanplex Free 4.0 is free forever for unlimited users. Standard is $49.9 per user per year and Premium is $69.9 per user per year. The pricing page has the details, and plans change, so check the pricing page before you decide.
You can run Sanplex in the cloud or on your own servers. The on-premises option can sit on an internal network isolated from the internet.
Where Sanplex does not fit
Based on the Agent guides, chat with project data works from the last synchronization, and data vectorization doesn't currently enforce Sanplex permissions. If you need live answers or strict, per user access control in chat, that matters. Agent features sit in the paid plans, not Sanplex Free 4.0. We haven't covered every Agent setting, so use the guides for the exact steps.
A worked example
This is an illustration, not a case study. A QA lead wants quick answers about open bugs before a release meeting.
-
Requirements. Answers about open bugs by feature, checked against the tracker.
-
Build. An admin connects ZAI, enables data vectorization and waits for synchronization.
-
Test. The QA lead asks ten questions using @Bug and checks each answer against the bug list.
-
Deploy. The team starts using it before reviews, with a rule to open the source item for any number they quote.
-
Maintain. Each month they review wrong answers and check the synchronization timing.
Writing prompts that use context well
Context only helps if the question uses it. A vague question wastes good data.
|
Weak prompt |
Better prompt |
Why it's better |
|---|---|---|
|
What's going on with bugs? |
Using @Bug, which bugs mention export and are still open? |
Names the data, the topic and the status |
|
Summarize the project |
Summarize @Requirement#123 and list what's unclear |
One object, one task |
|
Is the spec OK? |
Using @Current Document, list sections that contradict each other |
A clear, checkable job |
Ask for a list or a count only when you're prepared to verify it. For summaries and comparisons the chat is on firmer ground.
Starting fresh after big changes
Because context loads when the conversation begins, an old conversation may not see something you changed this morning. Start a new conversation after major updates, and keep one conversation per topic so the context stays focused.
Common mistakes to avoid
-
Treating chat as live. It reflects the last synchronization.
-
Skipping the permission check. Decide before you enable it.
-
Asking too broadly. A specific mention beats the whole project.
-
Quoting numbers unchecked. Verify counts.
-
Never starting a new conversation. Context loads at the start.
Chat or filters: which should you use?
Chat isn't a replacement for the tool's own filters and reports. When you know exactly what you want, a saved filter is quicker and always current. Chat earns its place when the question is fuzzy, spans several object types or needs a summary: "what are the main risks across these requirements?" or "which bugs relate to export?"
A useful habit is to use chat to find the area, then open the real items to confirm. You get the speed of chat and the certainty of the record.
It also helps to tell people what the chat can't see. If a decision lived in a meeting or a chat thread, the project data doesn't hold it, and the answer won't either.
What to vectorize first
You don't have to put everything in at once. A narrow start makes the permission question easier and the answers easier to check.
-
Start with documents everyone can read. Specs and process pages carry little risk.
-
Add bugs next. They're structured, so answers are easy to verify.
-
Hold back sensitive requirements. Wait until you've settled the permission question.
-
Review after a month. Widen scope only if the answers earned it.
Checklist before you start
-
ZAI connected and understood
-
Data vectorization scope decided
-
Permission impact reviewed
-
Synchronization timing known
-
Twenty test questions written
-
A rule for checking sources
-
A named owner
Try it on one project
Set up a project on Sanplex Free 4.0 first, then check the guides for agent options. The whole product is at sanplex.com.
Frequently asked questions
How do I chat with my project data?
You need an AI tool that can read your project data as context. In Sanplex, an admin connects ZAI, enables data vectorization, and users reference data in Agent chat with @mentions.
What are @mentions in Sanplex Agent chat?
They add context to a conversation. @Current Document uses the document you're viewing, @Sanplex uses all vectorized data, @Bug uses all bug data and @Requirement#123 uses one specific requirement.
Is chat with project data real time?
No. According to the Agent guides, context from vectorized data uses the data from the last synchronization and doesn't update in real time.
Does data vectorization respect user permissions?
The guides state it doesn't currently support permission control, so a standard user may see data in chat they couldn't open in Sanplex. Organizations with strict access rules should avoid enabling it.
How accurate are AI answers about project data?
They can be good for summaries and searches, and weaker on counts and recent changes. Always check the source item.
What is data vectorization?
It converts data such as requirements, bugs and documents into vectors, a searchable form, so the model can reference it in conversation.
Do I need ZAI to chat with project data in Sanplex?
The guides describe the Agent as relying on ZAI as a foundational service. Check the ZAI configuration guide for how it applies to your setup.
How much does Sanplex cost?
Sanplex Free 4.0 is free forever for unlimited users. Standard is $49.9 per user per year and Premium is $69.9 per user per year. Sanplex Agent is in Standard and Premium. Check the pricing page, since plans change.
Can the chat run on my own servers?
Sanplex can run on premises on an internal network isolated from the internet. Check the Agent and ZAI guides for how the AI services work in that setup.
Keep reading
|
Ready to give your chat some context? Read how Sanplex Agent chat works with your data: open the Agent chat guide. |
Resource
- Blog
- Customer stories
- FAQ
Support
- Book a Demo
- Email Us: [email protected]
2026-11-17 13:21:38
0