One Prompt, Every Workforce Bot
Workforce Bots Agent lets Ask AI find the right specialist bot for a question. HR Policy, SOP Helper, Recruiting Coach, Compliance Mentor — list them, search all their uploaded documents in one call, and forward the question to the bot that owns the answer. Each bot answers from its own KB, with web search as a fallback only.
Why Per-App Bots Fragment The Question Surface
A business with five specialist bots has five places to ask. Without a discovery layer, users have to know which bot owns which question before they even start. Workforce Bots Agent makes "I don't know which bot to ask" a one-prompt problem.
Users Don't Know Which Bot Owns Which Question
"Where do I file my safety incident report?" Is that the HR Policy bot? The Compliance Mentor? The Safety Hub bot? The user doesn't know — and the docs are in one of the three. Without a cross-bot search, they pick wrong, get a blank answer, and ask a human instead.
Specialist Bots Become Discoverability Dead-Ends
The Compliance Mentor has 42 carefully-curated documents and answers great compliance questions — when somebody actually opens it. Most users never do. The bot's investment stays unused because there's no surface that pulls them in.
Ask AI Doesn't Know What Bots Exist
A user asks Ask AI a compliance question. Ask AI doesn't know there's a Compliance Mentor bot with the answer pre-cached. It answers from generic platform context instead, and the bot's knowledge stays unused.
Per-Bot Answer Quality Stays Hidden
Each bot lives in its own surface; nobody compares bot-vs-bot answer quality, document coverage, or hand-off rates. The right central view ("which bots are pulling weight, which need more docs?") doesn't exist by default.
Bot Owners Update Their Document Sets And Nobody Tells The Users
The HR Policy bot just absorbed the new leave policy. Three employees asked the question yesterday and got the old answer because nobody knew the refresh happened. A "what changed in the bot knowledge base this week?" rollup would have closed the gap before the wrong answers shipped.
Routing A Question To The Right Bot Costs A User Three Tries
The user asks the wrong bot, gets a "I don't have that info" reply, asks the next bot, gets pointed somewhere else, asks the third. By the third try the user gives up and pings a human. Discovery has to happen before the first ask, not after the third failure.
Workforce Bots Agent At A Glance
Workforce Bots AI
Discover and query your configured Workforce Bots. Cross-bot search, scoped routing.
Inside Workforce Bots Agent — The Actual Capabilities
Every block below maps to a real tool in Agents::ToolRegistry::WorkforceBots — the platform-level registry that Ask AI dispatches against. Each individual bot runs its own agentic loop using WorkforceBot (singular) tools — KB search first, web search only on miss, recommend-human-contact when both fail.
List Every Bot The Business Has Configured
list_workforce_bots answers "what bots do we have?" with the full inventory — bot name, template type, description, welcome message, and whether web search is enabled. Used directly when a user asks for the list, and as the routing prelude when the agent needs to pick a target for ask_workforce_bot.
- List configured bots — list_workforce_bots returns every Workforce Bot the business has set up.
- 5 template types — hr_policy, recruiting, process, compliance, custom.
- Welcome message included — surfaces the bot's user-facing intro so users know what each bot is for.
- Default limit 20 — businesses with more bots paginate.
Search Across Every Bot's Documents In One Call
search_workforce_bot_documents runs against every Workforce Bot's uploaded docs in a single call — returning matching chunks plus the bot that owns each match. The "I don't know which bot to ask" problem disappears: search across all of them, see which bot's KB has the answer, then forward to that bot.
- Search across all bot KBs — search_workforce_bot_documents covers every Workforce Bot's uploaded documents in one query.
- Returns the owning bot per chunk — each result is tagged with the bot that owns the document; the agent knows where to forward.
- Up to 10 results — limit (default 5, max 10) keeps the result set tight.
- Composable with ask_workforce_bot — search to find the owning bot, then forward the user's full question for the bot's grounded answer.
Forward To The Right Bot · Bot Runs Its Own Grounded Loop
Once the right bot is identified, ask_workforce_bot forwards the question to that bot. The bot then runs its own agentic loop using WorkforceBot tools — search_knowledge_base against its own docs first, web_search only if the bot has it enabled and the KB returned nothing, and recommend_human_contact as the final fallback.
- Forward to a specific bot — ask_workforce_bot by bot_id or bot_name; the question is passed verbatim.
- Each bot runs search_knowledge_base first — KB grounding is mandatory; the bot's answer must be grounded in its uploaded docs.
- Web search is the fallback — web_search runs only when the KB returns nothing AND the bot has web search enabled by the admin.
- Human escalation as universal extra — recommend_human_contact is the bot's last resort; returns the configured human contact for that bot.
Outcomes Teams Can Measure
Workforce Bots Agent's job is to make every specialist bot reachable from the same prompt surface and to put bot KB coverage on a measurable footing. Measure against the pre-agent baseline so you can see how the discovery layer changed bot usage.
- Per-bot usage — invocations via the discovery agent vs invocations via the bot's standalone surface; signals whether the cross-bot layer is pulling its weight.
- Cross-bot routing accuracy — share of search_workforce_bot_documents results that the user actually accepts as the right bot to forward to.
- KB-hit rate per bot — share of bot answers grounded in its own KB vs requiring web search or human escalation; canary for KB coverage gaps.
- Web-search fallback rate — how often bots reach for web search; high rates mean the KB is underfed.
- Recommend-human rate — share of bot conversations that end at human escalation; the metric every specialist bot owner wants to drive down.
Intentionally Read-Only · Bot Writes Live In The Bot's Own Surface
Workforce Bots Agent's RISKY_TOOLS list is empty. Three discovery tools, all reads — list bots, search across bot docs, forward a question. The agent never edits a bot's KB, never changes a welcome message, never reconfigures web-search or human-escalation contacts. Those writes live in the bot admin surface where each bot's owning team makes the call.
- RISKY_TOOLS is empty — 3 read tools; zero writes.
- Per-bot writes stay in the bot admin UI — KB uploads, welcome message edits, web search toggle, human contact configuration all live in each bot's own admin surface.
- Each bot's answer loop is grounded — WorkforceBot (singular) tools require search_knowledge_base first; bots can't answer from generic LLM knowledge without grounding.
- Audit trail per forwarded question — ask_workforce_bot calls log the requesting user, the target bot, and the question forwarded.
WHAT TEAMS TRY INSTEAD
The four alternatives — and none of them route to the right specialist while keeping each bot grounded in its own KB
Specialist bots multiply quickly — one HR policy bot, one SOP bot, one recruiting coach — and users have no idea which to ask. None of the alternatives discover the right bot, forward the question verbatim, and keep each bot answering from its own KB.
ChatGPT, Claude, or Copilot as a single generalist
One generic AI guessing across HR, SOP, compliance, and recruiting
- Workforce Bots Agent forwards to the specialist bot grounded in the actual HR / SOP / compliance KB the team curates
- Each bot's answer is KB-grounded; ChatGPT answers from training data with no source attribution
- Cross-bot document search runs across every uploaded doc in one call — generic AI has zero visibility into your KBs
Zapier AI, Workato Copilot, Power Automate Copilot
Workflow-automation AI routing requests via integrations
- Bots are domain-curated KBs with welcome messages and recommend-human contacts — not generic webhook triggers
- One forwarded question loops through the right bot's agentic answer engine with KB attribution — Zapier shuttles strings
- No third-party automation service holds your KB content; everything stays inside MangoApps
Custom router service in front of multiple LangChain bots
An AI platform team's bespoke routing layer, then maintenance
- Already shipped — bot templates (hr_policy / recruiting / process / compliance / custom), KB ingestion, routing, audit, and admin surface in place
- Two-step discovery (list_workforce_bots when named, cross-bot doc search when unknown) — no custom intent classifier to retrain
- When a new bot template arrives (e.g. compliance_mentor), the discovery agent picks it up — no router rewrite
"Just have a #ask-hr Slack channel and rotate the on-call"
The status quo — humans triaging generalist questions
- One Ask AI prompt finds the right specialist bot instead of a channel ping that interrupts whoever's on rotation
- Each bot's KB-hit rate is measurable; channel rotations have no measurement
- recommend-human-contact stays as the fallback when the bot's KB doesn't cover a topic — no Slack noise for things the KB handles
PLATFORM ADVANTAGE
Workforce Bots Agent inherits everything the platform already runs
A custom router service has to plumb each of these. The discovery agent gets them for free.
Two-step discovery
list_workforce_bots when the user names a bot directly; search_workforce_bot_documents when the right bot is unknown — the search result names the target.
KB-grounded per bot
Each WorkforceBot answer requires search_knowledge_base first — bots can't free-style from generic LLM knowledge.
Five bot templates
hr_policy, recruiting, process, compliance, custom — each with the right starter toolset (HR Policy includes employee directory lookup).
Web search as documented fallback
When the KB doesn't cover a topic, the bot can fall back to web search — explicitly, with the source surfaced.
recommend-human-contact
Every bot has a configured human escalation — the bot can route to a real coworker when the question is out of scope.
RubyLLM model tiering
Discovery and KB search run on small tier; specialist-bot answer loops on standard. Per-conversation cost stays low at scale.
INDUSTRY FIT
Industries where specialist-bot discovery moves the most weight
Workforce Bots Agent shines where domain knowledge is segmented (HR vs SOP vs compliance) and users don't know whose KB to ask first.
Healthcare
Clinical SOP, HR policy, and credentialing bots discoverable from one prompt — clinicians don't navigate three portals during a shift.
Manufacturing
Safety bot, quality bot, and HR bot routed automatically from a single Ask AI question — operators get the right answer without naming the bot.
Retail
Product-knowledge, brand-standards, and policy bots discoverable per associate; managers don't field every "how do I do X" question.
Financial Services
Compliance, HR, and product bots with grounded KBs; one prompt forwards to the right one without leaking pre-trade questions to the wrong domain.
Public Sector
Multi-bureau specialist bots discoverable across the agency; within the FedRAMP-eligible boundary, no third-party routing SaaS.
Professional Services
Methodology, sales-process, and compliance bots routed for consultants — the right specialist on the first prompt, not a Slack roundtrip.
WHY MANGOAPPS WINS
An embedded discovery agent beats a generalist, a Zap router, or a custom build on every axis
The argument bot owners, IT, security, and the user typing the question all share — and the one a horizontal AI or a workflow-automation copilot structurally cannot answer.
Cheaper than the alternatives
No Zapier seat per bot, no Workato copilot license, no custom router-service build, no extra HR/compliance headcount answering "which bot do I ask".
More secure
Read-only discovery — zero write tools, every forwarded question logged through AiApiLog, KB content stays inside the tenant.
Easier to deploy
Already deployed once a bot exists. The discovery agent lights up the moment the second specialist bot is configured.
Easier to use
One Ask AI prompt finds the bot, forwards the question, and surfaces the bot's KB-grounded answer — no bot-picker UI to learn.
Easier to manage
Per-bot KB ingestion, welcome message, web-search toggle, and human-contact escalation all live in each bot's admin surface — one console, many bots.
Easier to extend
A new bot template (e.g. benefits_mentor) adds a tool set automatically — the discovery agent picks it up without retraining.
AI is actually better
A generalist or workflow AI can answer one thing. Only Workforce Bots Agent can also discover the right specialist, forward the question verbatim, ground the answer in the right KB, and fall back to a human contact when the KB doesn't cover it.
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 Workforce Bots Agent
3 tools — list_workforce_bots (full inventory of configured bots), search_workforce_bot_documents (cross-bot document search returning the owning bot per chunk), and ask_workforce_bot (forward a question to a specific bot by id or name). All read-only.
The singular WorkforceBot agent runs an individual bot's agentic loop — search_knowledge_base + web_search + recommend_human_contact — using only that bot's uploaded documents. The plural Workforce Bots Agent is the platform-level discovery layer that Ask AI dispatches against to FIND the right bot, then forwards a question to it. Distinct registry, distinct purpose.
Five — hr_policy, recruiting, process, compliance, custom. The HR Policy template includes a lookup_employee_directory tool extra so the bot can point users to a real coworker for follow-up. Other templates inherit the universal toolset (KB search, web search, recommend-human-contact) without extras.
Two-step. list_workforce_bots when the user names a bot directly, or search_workforce_bot_documents when the right bot is unknown — the search returns the owning bot per chunk, so the highest-scoring result names the target. The ask_workforce_bot call then forwards the question verbatim to that bot's agentic loop.
Per-bot usage (discovery agent vs standalone surface), cross-bot routing accuracy, KB-hit rate per bot, web-search fallback rate, and recommend-human rate. The KB-hit rate is the single most important canary — when it drops, the bot's KB needs more documents.
Let's Talk
Since 2008, we've been building the employee platform for the frontline, earning the trust of 2 million+ users and an NPS of 78.
Why Choose Us?
- AI-Ready Platform: One intelligent place for every employee and workflow.
- Top Security: HITRUST, ISO & SOC 2 certified.
- Exceptional UX: Delightful on mobile and desktop.
- Proven Results: 98% customer retention rate.
Trusted by Legendary Companies:
Prefer to explore first? Ask AI about Workforce Bots AI Agent →