Drive Files In Seconds, Not Searches
Five tools for finding things in Drive — by filename keyword, by content keyword, by recent activity, by who shared it, or by natural-language description. Plus a summarize-this-file tool for PDFs and text. Strictly read-only — no creates, no shares, no deletes. Drive Admin Agent handles those.
Why "Find My File" Is Still Five Clicks
Drive Help Agent fixes all four — keyword search, semantic description search, recent and shared lists, and one-file summaries — all read-only, all permission-aware.
Filename Recall Is A Coin Flip
Was the file called Q3-Budget-Forecast or Budget-Forecast-Q3 or FY26-Q3-Plan-rev3? Filename search punishes the slightest typo and rewards only the originator who named it.
"That File About The Office Layout" Has No Filename
Users remember concepts, not filenames. A keyword search misses every file that doesn't literally use the word — even when the content matches the idea perfectly.
Shared-With-Me Is A Mixing Bowl
Everyone shares everything, the shared-with-me view becomes a 400-file pile, and the file you actually need from yesterday is buried behind newsletters and dashboard invites you opened once.
Reading A 14-Page PDF To Find One Sentence
The security policy is 14 pages. The question is "how often do we review access?" A keyword search returns the doc; the user still has to open it, scroll, and ctrl-F. The 30-second answer takes five minutes.
Duplicate Files Make Picking The Right Version Impossible
The same policy exists as policy-v1.pdf, policy-final.pdf, and policy-final-FINAL.pdf — sometimes in three different folders. Search returns all three; nobody knows which is current. The user picks based on the most recent modified date and finds out two days later they were reading a draft a coworker forgot to clean up.
"Who Sent Me That File Last Week?" Has No Easy Answer
The user knows it landed in a chat or an email seven days ago, and the sender was either a manager or a teammate from a related team — but they didn't bookmark it. Without a fast "files shared with me by person, in this timeframe" lookup, finding the doc means scrolling back through chat history and inbox threads for half an hour.
Drive Help Agent At A Glance
Drive AI (Help)
Five read-only ways to find a file. Plus summaries.
Inside Drive Help Agent — The Actual Capabilities
Every block below maps to a real tool the agent uses against your Drive. All five tools are read-only — the agent finds, lists, and summarizes; it never creates, shares, or deletes. Writes belong to Drive Admin Agent.
Find Files By Filename Or Content Keyword
The classic "I know roughly what it's called" path. The agent searches filenames and extracted file content together, returns ranked matches with snippets, and respects the user's existing access.
- Search filenames and content via search_drive — keyword or content phrase, ranked matches.
- Snippet on every result — see where the term appears before opening the file.
- Limit configurable — default 10 results, max 25.
- Permission-aware — the agent only returns files the user can already open.
Recently Touched And Shared With Me — Two Lists, One Glance
The two most common Drive entry points — "what did I work on yesterday?" and "what did Mira share with me?" — answered as discrete lists. No more digging through one cluttered shared-with-me view.
- List my recently touched files via list_recent_files — files I opened or edited.
- List files shared with me via list_shared_with_me — files others have shared, with sharer attribution.
- Limit per list — default 10, max 25 per call.
- Distinct from search — these don't accept a keyword; they're the "show me what's near me" entry points.
Semantic Search When You Don't Know The Filename
"Find the doc that talks about Q3 revenue targets" or "the PDF where we agreed on the office layout" — described, not named. Semantic embedding search returns the most likely match without requiring keyword overlap.
- Semantic embedding search via find_file_by_description — describe the content, not the filename.
- Returns top matches — default 5, max 15.
- Complements keyword search — when filenames are forgotten, semantic recall catches what search_drive misses.
- Permission-aware — embeddings only cover files the user can already see.
TL;DR A Specific File — PDF Or Text
Drop a filename or DriveItem ID and the agent returns a structured summary of the extracted text. Works for PDFs and plain-text files where content has been indexed. Useful for policy docs, contracts, design rationale.
- Summarize a single file via summarize_file — by DriveItem ID (preferred) or filename.
- Works on extracted text — PDFs and plain text where content has already been indexed by the Drive ingestion pipeline.
- Grounded in the file — the summary is built from the actual extracted content, not the file's metadata.
- Strictly read — the summary doesn't modify the file or its extracted text store.
Outcomes Teams Can Measure
The agent compresses Drive search and summarization into one chat ask. Measure against your pre-agent baseline.
- Time to find a file — seconds from "where's the Q3 budget?" to the opened file.
- Semantic-find usage — share of finds done by description vs filename keyword.
- "Where's this file?" Slack messages — declining peer requests as users self-serve.
- File summaries opened vs full reads — share of policy / contract questions answered by summary alone.
- Cross-team retrieval — files surfaced from shared collaborators without owner intervention.
Intentionally Read-Only — Zero Write Tools
Drive Help Agent has 5 tools — all read. There is no create_folder, no share_drive_item, no trash_drive_item. Those are Drive Admin Agent's job. This split keeps Drive Help safe to enable broadly while Drive Admin stays gated by role + confirmation.
- Zero write tools — the agent's RISKY_TOOLS list is empty. No creates, no shares, no deletes.
- Permission-aware — every read goes through the Drive access model; nothing crosses team boundaries.
- Paired with Drive Admin — read here, write there; the split is intentional.
- Audit trail on every read — every tool call logs the user, the tool, and the parameters.
WHAT TEAMS TRY INSTEAD
The four alternatives — and why none of them find the file or respect your scope
When a user can't find the file, they reach for one of these four. None do semantic description search across the actual files the user has access to, with the source cited and the read logged.
Pasting a file's content into ChatGPT, Claude, or Copilot
General-purpose AI summarizing pasted text
- Drive Help Agent finds the file first — pasting requires the user to already have the file open
- Semantic search matches "the office layout doc" to files whose filename doesn't include those words
- Every read logs to AiApiLog with the requester and the file ID — no shadow-IT data exfil
Box AI, Google Drive Duet AI, OneDrive Copilot
Vendor-trapped search AI inside a single storage silo
- Searches the same Drive workspace where Wikis, Libraries, and HR Files also live — one chat surface, not three vendor searches
- Honors the MangoApps permission model on every read; vendor AI has its own ACL that drifts
- Works for frontline employees who never had a Box/Drive seat in the first place
Custom RAG over a shared drive
An engineering team's six-month build, then forever maintenance
- Shipped already — engineering doesn't need to plumb identity, audit, retention, or semantic embeddings
- Permission-aware on every retrieval; most DIY builds collapse permission at index time and leak access
- Inherits new file types and new tools automatically; DIY indexes are frozen on the format they knew at launch
The manual fallback — "ask whoever sent it"
The default when AI tools fall short
- Files-shared-with-me-by-person queries return the answer in seconds instead of half an hour scrolling chat history
- One-file summary returns the answer to "how often do we review access?" without anyone reading 14 pages
- Frees senders from re-sharing files to teammates who lost the original link
PLATFORM LEVERAGE
Drive Help Agent inherits everything the platform already runs
A custom-built file-search bot has to plumb each of these. Drive Help Agent gets them for free.
Cross-app data plane
Knows that a file lives in a Workspace, attaches to a Wiki, or is governed by HR Files — so search results carry the right context, not just a filename.
Source-app permission enforcement
Every read goes through Drive's access model. The agent cannot return a file the user couldn't open directly — even if the embedding index would have surfaced it.
Audit trail & retention
Even read-only searches log to AiApiLog with the requesting user, the query, and the file IDs returned. Audits stop having gaps.
Translation in 100+ languages
A user in Mexico asks "encuentra la política de vacaciones" — the semantic search returns the English policy and the agent summarizes it in Spanish.
Mobile-first search
Frontline employees find policies and shift documents from the same mobile app they already opened for shifts and pay.
RubyLLM-grounded model tiering
Filename search runs on nano; semantic description search and PDF summaries route up. Routing is automatic and per call.
INDUSTRY FIT
Industries where finding-the-file fast moves the most weight
Drive Help Agent shines wherever the file exists, the user knows it exists, and the filename has been forgotten.
Healthcare
Charge nurse asks "find the latest infection-control SOP" — semantic search returns the current version, not the three superseded copies still in the folder.
Manufacturing
Plant supervisor asks for the calibration procedure on press 7 — agent surfaces the file with the inspector who last updated it and the date.
Retail
Store managers find brand-standard merchandising files by concept ("holiday endcap layout") instead of trying to recall the filename HQ used.
Professional Services
Consultants find the right precedent deck or proposal template in seconds across hundreds of client folders — semantic match, scoped to their permissions.
Financial Services
Compliance officers locate the version of a procedure that was in force on a specific date — search returns the right file with last-updated timestamp cited.
Public Sector
FedRAMP-eligible deployment options keep every search inside the tenant boundary with full audit logging on read access.
WHY MANGOAPPS WINS
An embedded file-search agent beats a storage-vendor copilot, a horizontal chatbot, or a DIY index on every axis
The argument end users, IT, security, and operations all share — and the one a single-vendor storage AI structurally cannot answer.
Cheaper than the alternatives
No per-seat storage-vendor copilot, no per-seat ChatGPT license, no six-month RAG build, no internal helpdesk hours soaked by "where's that file" tickets.
More secure
Read-only by design. Every retrieval respects Drive's permission model and logs to AiApiLog. Nothing leaves the tenant boundary.
Easier to deploy
Already deployed if Drive is enabled. No embedding pipeline to build, no vector store to operate, no ACL bridge to maintain.
Easier to use
One chat surface for filename, content, semantic, recent, shared, and one-file summary — no five different search UIs.
Easier to manage
Permission changes happen once in Drive and the agent inherits them — no parallel ACL or stale-index drift.
Easier to extend
New file types and new app surfaces become searchable the day they ship — no embedding-pipeline refresh.
AI is actually better
A vendor copilot finds files inside one product. Only Drive Help Agent matches a concept across Drive, Wikis, Libraries, and HR Files in one query — and respects the user's exact scope on each surface.
At our head office within HR, marketing, and finance, we use ChatGPT quite a bit, and MangoApps AI Helper is a useful tool for us. As I go around to the stores, I’ve been showing them how the AI Intranet Helper can find things and answer plain language questions about our resources, and they’re quite excited about that.
Alison van Zijl
Head of Human Resources
Full House
Customer Success
Related Customer Stories
Customer Case Studies
Video Case Studies
Customer Case Studies
Customer Case Studies
Customer Case Studies
Customer Case Studies
Frequently Asked Questions About Drive Help Agent
5 read-only tools — search Drive by filename or content keyword (search_drive), list recently touched files (list_recent_files), list files shared with me (list_shared_with_me), find a file by natural-language description with semantic embedding search (find_file_by_description), and summarize a specific file's extracted text (summarize_file).
No. Drive Help Agent is strictly read-only — its RISKY_TOOLS list is empty. It cannot create folders, share files, revoke access, or delete anything. Writes are handled by Drive Admin Agent.
search_drive matches filenames and content keywords ranked by relevance — best when the user remembers approximate terms in the file. find_file_by_description uses semantic embedding search — best when the user describes content concepts without specific keywords.
summarize_file works on files where the Drive ingestion pipeline has already extracted text — primarily PDFs and plain text. Files without extracted content (raw images, binary formats) won't have a useful summary.
Time to find a file, semantic-find usage (description vs keyword), declining "where is this?" Slack messages, file summaries opened vs full reads, and cross-team retrieval where users find files without pinging the owner. Compare against your pre-agent baseline.
Let's Talk
Since 2008, we've been building the workforce platform — earning the trust of 2 million+ users and an NPS of 78.
Why Choose Us?
- AI-Powered Platform: The most unified workforce experience on the planet.
- Top Security: HITRUST, ISO & SOC 2 certified.
- Exceptional UX: Delightful on mobile and desktop.
- Proven Results: 98% customer retention rate.
Trusted by Legendary Companies:
