Knowledge Base Chat
Answer questions from the app's approved help articles and policies, with sources shown.
What it adds
A retrieval-backed answering surface over the app's published knowledge sources, restricted to what the asker is allowed to read.
What your agent is told to do
5
What your agent is told to do
5-
1
Define which collections count as knowledge sources and mark each with an audience — public, customer, or internal. Filter the retrievable set by the asker's audience before retrieval, not after the answer is written.
-
2
Restrict retrieval to currently published versions. Drafts, archived articles, and superseded revisions must be excluded from the corpus, not merely ranked lower.
-
3
Cite every source used, and when two retrieved sources contradict each other, present both with their dates rather than picking one and hiding the conflict.
-
4
Keep the answer inside the retrieved evidence. If retrieval returns nothing relevant, say the knowledge base does not cover it and offer the search results or a route to support.
-
5
Re-index a source when it is published, edited, unpublished, or has its audience changed, using the app's existing background-job system, and drop any cached answer that cited it.
Edge cases it handles
8
Edge cases it handles
8- An internal policy retrieved for a customer question is a disclosure incident. Audience filtering must run inside the retrieval query so no restricted passage ever reaches the model.
- A draft rewrite of an article must never outrank the published version. If a source has no published revision, it is not in the corpus at all.
- Contradictory sources are common when an old article was never retired. Show both with publication dates and say which is more recent rather than silently averaging them.
- Answering beyond the retrieved evidence produces plausible policy that does not exist. Anything the retrieved passages do not support must be omitted, not softened with a hedge.
- Edits must invalidate cached answers immediately. Serving yesterday's answer with today's citation is worse than having no cache, because the citation looks like verification.
- Show the answer as assistance, not authority, and keep the underlying articles one click away so a user can check it.
- When the model is unavailable, fall back to the existing keyword search over the same articles rather than showing an empty page.
- Document Question Answering covers a single file uploaded by one user; this feature covers the app's shared published sources. Keep the two corpora and their permission models separate.
Definition of done
8
Definition of done
8- Retrieval is filtered by the asker's audience before any passage reaches the model.
- Only currently published revisions are retrievable; drafts and archived material are excluded.
- Every answer lists its sources, and conflicting sources are shown together with dates.
- Questions unsupported by retrieved evidence return an explicit no-answer response.
- Publishing, editing, or unpublishing a source re-indexes it and invalidates answers that cited it.
- Keyword search over the same articles remains available when the model is unavailable.
- The feature matches the existing design system.
- No existing functionality is broken.
Related features
Natural Language Filters
Natural Language Filters
Let people describe what they want to see and turn it into filters they can inspect and edit.
What it does
A plain-language input on list views that produces the app's normal filter chips, shown for review before they apply.
How it works
- 1 Give the model a machine-readable description of the filterable fields, their types, their allowed operators, and their permitted values, derived from the app's existing filter definitions. It selects from that description and returns a structured filter set; it does not write query syntax.
- 2 Render the result as the app's ordinary filter controls, in the pending state, and require the user to apply them. Do not run the query on the model's output directly, and do not skip the review step even when confidence looks high.
- 3 Resolve relative expressions such as last week or this quarter against the user's time zone and the workspace's week and fiscal-year settings, and show the resolved absolute dates in the chip so the user can see what was assumed.
Copy the prompt
No account needed
Add this feature to my app:
https://addthisfeature.com/x/natural-language-filters
AI Insight Cards
AI Insight Cards
Turn a report into a few grounded cards, each tied to a number the app calculated.
What it does
A small set of cards summarising notable movements in a report, each linked to the query and figure that produced it.
How it works
- 1 Detect candidate movements in the app first, using the report's own aggregations. The model ranks and phrases the candidates it is given; it does not go looking for them and it does not produce the arithmetic.
- 2 Bind each card to the query, metric, and period that support it, and make the card link through to the filtered view so a reader can check it in one click.
- 3 Set a materiality threshold before generation, in both relative and absolute terms, so a swing on a metric with three events does not outrank a real change on a metric with thousands.
Copy the prompt
No account needed
Add this feature to my app:
https://addthisfeature.com/x/ai-insight-cards
AI Trend Analysis
AI Trend Analysis
Explain what a chart is actually showing, in sentences backed by computed numbers.
What it does
A written summary attached to a time series or metric set, describing direction, magnitude, and comparison period.
How it works
- 1 Compute every number in ordinary code before the model is involved: the change, the rate, the comparison period, the baseline, the seasonal adjustment. The model writes prose about figures it is given and must never produce a figure of its own.
- 2 State the comparison window and the time zone in the summary text, resolved from the workspace's configured zone rather than the server's. A change described as week over week is meaningless without saying which weeks.
- 3 Instruct the model to describe what the data shows and to stop there. Do not let it assert causes, attribute movement to campaigns or releases, or predict what happens next.
Copy the prompt
No account needed
Add this feature to my app:
https://addthisfeature.com/x/ai-trend-analysis
How it works
-
1
Copy the link
Grab the Markdown instruction URL for this feature.
-
2
Give it to your AI
Paste it into Claude Code, Cursor, v0, Lovable — whatever you build with.
-
3
It inspects, then implements
Your agent reads your existing app first, then adds the feature to fit it.
Works with your stack
These instructions are written to adapt. They tell the agent to detect your framework, match your existing design system, and reuse what you already have — rather than assuming a particular stack.
Need it tighter than that? Customize the feature and tell it exactly what you're running.