Ingests your docs
Pulls in drives, wikis, PDFs, and past tickets and organises them: no manual re-authoring of what you already wrote.
Task tier · Knowledge
An AI agent that turns your scattered docs into answers.
It ingests the documents you already have, organises them into a searchable knowledge base, answers questions against it for your team or customers, and keeps it current by spotting gaps and stale content as things change.
ModelMix of LLMs + retrieval (RAG)
Cost per taskModel usage billed at cost, measured per task in your pilot
Connects toGoogle DriveSharePointNotionConfluenceSlack
Autonomy
- Suggests
- Acts with approval
- Acts on its own
Answers from approved sources only, with role-based access, and flags gaps and stale pages to an owner. How autonomy works
What does an AI knowledge base agent do?
A knowledge base agent builds and maintains the source of truth other agents and people rely on. It ingests your existing documents (drives, wikis, PDFs, tickets), structures them into a searchable knowledge base, answers natural-language questions against it for staff or customers, and maintains it, flagging gaps from unanswered questions and stale content as your business changes. Task tier, running on Workers AI in your own Cloudflare account, with retrieval. It's the foundation that powers a chatbot, a support agent, or a call answering agent.
theagency47 · Updated June 2026 The spec sheet
How the Knowledge Base Agent actually works.
| Job-to-be-done | Turn scattered documents into a searchable, maintained knowledge base that answers questions |
| Tier | Task (bounded, high volume) |
| Underlying model | Workers AI models in your own Cloudflare account + retrieval (RAG) |
| Trigger | On query (answers) + scheduled ingest/freshness sweep |
| Inputs | Source docs (Drive, wiki, PDFs, tickets), access rules, taxonomy, freshness policy |
| Tools available | Document ingest, vector index, search, gap detection, access control |
| Autonomous decisions | Chunking/structuring, answer assembly, gap + staleness flagging |
| Escalation rules | No supporting content → "not documented" + flag gap · Conflicting docs → surface both + flag · Restricted content → access-gated |
| KPIs measured | Coverage, answer rate, citation accuracy, gaps found/closed, staleness rate |
| Eval suite | 20 test cases (citation grounding, gap detection, access enforcement). |
What it does
The work it takes off your team.
Answers with citations
Replies to questions from the source content and cites where each answer came from, so it's verifiable.
Finds the gaps
Tracks questions it can't answer and surfaces exactly what documentation is missing.
Flags stale content
Spots content that's aged or conflicts with newer material so the base stays trustworthy.
Access-aware
Honours who can see what: internal vs customer-facing, team-scoped: so one base serves multiple audiences safely.
Powers other agents
Acts as the grounded source for chatbots, support, and voice agents, so they all answer consistently.
What it looks like in practice
A typical run.
- Continuously, Staff and customers ask questions and get cited answers from the latest content.
- On ingest, New and updated documents are pulled in, structured, and indexed automatically.
- Freshness sweep, Stale or conflicting content is flagged for an owner to review.
- Weekly, Gap report: top unanswered questions and what to document next.
FAQ
Questions about the Knowledge Base Agent.
Do I have to rewrite our documentation?
No, it works from what you already have (drives, wikis, tickets, PDFs). It structures and indexes existing content, and tells you where the real gaps are so you write only what's missing.
Can it serve both staff and customers?
Yes. Access rules let one base answer internal questions for the team and a scoped subset for customers, without leaking internal-only content.
How does it relate to the chatbot and support agents?
It's the foundation. A chatbot, support agent, or call answering agent all answer from this base, so build it once and every customer-facing agent gets smarter.
Related
Agents and engagements that pair with this.
Related: See the chatbot →
