Knowledge Base

The Knowledge Base is where you give your voice agents the documents they can read from during live calls. Instead of relying only on the instructions you write, an agent can look up real facts — your price list, opening hours, policies, product details — and answer callers accurately and consistently. Upload a file once, mark it global or assign it to specific agents, and it becomes part of what your agents "know".

In this guide

• What the Knowledge Base is
• Opening the Knowledge Base
• Tour of the screen
• Supported file types
• Uploading a document
• Processing and indexing status
• Browsing and reading documents
• Editing a document's content
• Controlling availability: global vs per-agent
• Assigning documents to agents
• Attaching documents from the agent editor
• How an agent uses the Knowledge Base during a call
• Retrying and fixing stuck or failed documents
• Deleting a document
• Best practices for great Knowledge Base content
• Common workflows
• Troubleshooting
• Quick reference

---

What the Knowledge Base is

A Knowledge Base is a library of documents that your voice agents can search and reference while they are talking to a caller. Think of it as the agent's reference shelf.

When a caller asks a question, the agent doesn't just guess from its general instructions — it can pull the most relevant passages out of your uploaded documents and use them to give a precise, grounded answer. This is what keeps answers accurate: prices come from your price list, your refund window comes from your policy document, and so on.

A few key ideas to keep in mind:

• Documents belong to your organisation. Everything you upload lives in your account's Knowledge Base and is shared across the agents you choose.
• Each document can be global or per-agent. A global document is automatically available to every agent. A per-agent document is only available to the specific agents you assign it to.
• Documents are processed before they can be used. After upload, the platform reads, splits, and indexes the file so agents can search inside it. A document only becomes usable once it shows the Ready status.
• You stay in control. You can preview content, edit text-based files in place, reassign documents, and delete anything you no longer need — all from one screen.

---

Opening the Knowledge Base

1. Open the Voice AI section of your portal.
2. In the left-hand navigation, click Knowledge Base (the open-book icon).

The page header reads Knowledge Base with the subtitle "Documents your agents retrieve during calls for accurate, grounded responses."

The Knowledge Base is included on every Voice AI plan, so it is always available once Voice AI is enabled for your account. If you ever see an "upgrade" or "not enabled" screen instead, your account hasn't had Voice AI switched on yet — contact your administrator.

---

Tour of the screen

The Knowledge Base screen has two main parts:

The document list uses these columns:

The two action buttons on the right appear when you hover over a row. The Global toggle stays visible if a document is already global.

If you have not uploaded anything yet, the list is replaced by an empty state that reads "No documents yet — Give your agent knowledge to draw on during calls." with an Upload Document button.

---

Supported file types

You can upload the following file types. They fall into three groups based on how the platform handles them.

The file picker is restricted to these types (.pdf, .txt, .docx, .md, .csv, .png, .jpg, .jpeg, .webp, .json), so you won't be able to select anything else.

Tip: For knowledge you want the agent to quote accurately — prices, policies, FAQs — plain text (.txt, .md) and CSV are the most reliable formats, because they're clean and you can edit them in place without re-uploading.

---

Uploading a document

You can add a document in two ways.

Method 1: The Upload button

1. Click Upload in the top-right of the page.
2. Choose a file from your computer in the file picker.
3. The upload starts immediately.

Method 2: Drag and drop

1. Drag a file from your desktop onto the Knowledge Base page.
2. A full-screen overlay appears reading "Drop to upload".
3. Release the file to begin the upload.

What happens during upload

While the file uploads, a progress bar appears below the header showing "Uploading…" and a live percentage. The upload runs in stages:

1. The file is securely transferred to storage.
2. The platform queues the document for processing.
3. You see a confirmation: ""filename" uploaded — processing started."

Once that toast appears, the new document shows up in the list with a Pending or Processing status. You can keep working — uploads and processing happen in the background.

You upload one file at a time. To add several documents, repeat the upload for each file. New documents are always created in chunked retrieval mode (see below).

---

Processing and indexing status

After you upload a file, the platform reads it, splits it into searchable pieces (called chunks), and indexes those pieces so an agent can look things up quickly. This takes a moment — larger files take longer.

Every document shows one of four statuses:

A document can only be attached to and used by agents once it is Ready. While a document is Pending or Processing, it appears in the agent's document list but cannot be switched on.

Live updates

You don't need to refresh manually. While any document is Pending or Processing, the page automatically checks for updates every few seconds and moves documents to Ready (or Failed) as soon as the platform finishes. You can also click Refresh at any time to reload the list immediately.

Chunks

When a document is Ready, the list shows how many chunks it was split into (for example, 12 chunks). Each chunk is a searchable passage. More chunks generally means more content — and more places the agent can pull an answer from. You'll also see the chunk count on the Details tab.

---

Browsing and reading documents

Click any row to open the detail panel on the right. It has a header (with the filename and action buttons) and three tabs.

The Content tab

The Content tab shows the document so you can confirm what your agents will be working from. What you see depends on the file type:

• Text, Markdown, CSV, JSON — the text is shown inline, and you can edit it (see the next section).
• PDF — shown in an embedded viewer, with Open (new tab) and Download buttons.
• Images — shown as a preview.
• Other / unsupported preview — a placeholder with the filename and an Open file button.

The panel header also has its own quick actions: open in a new tab (↗), Download (⬇), Delete (🗑), and Close (✕).

The Details tab

The Details tab lists the document's properties:

If the document is still processing or has failed, this tab also shows a coloured panel explaining the situation, with a link to retry.

The Agents tab

The Agents tab controls which agents can use the document. It's covered in detail below.

---

Editing a document's content

For text-based files — .txt, .md, .csv, .json, and HTML — you can edit the content directly in the panel without re-uploading. This is the fastest way to keep your knowledge current.

1. Open the document and go to the Content tab.
2. Click Edit (the pencil) in the toolbar.
3. The content becomes an editable text box, and the toolbar shows an Editing label.
4. Make your changes.
5. Click Save & Reprocess to save, or Cancel to discard.

When you save, the platform stores the new text and automatically re-processes the document so the agent's searchable index reflects your edits. You'll see "Saved — document queued for reprocessing," and the status returns to processing briefly before going back to Ready.

Note: PDFs, Word documents, and images cannot be edited in place — there's no Edit button for them. To change those, update the source file and upload it again, then delete the old version.

---

Controlling availability: global vs per-agent

Every document is either global or per-agent. This decides which agents can use it.

A global document shows a Global pill next to its name in the list.

Making a document global (or per-agent)

There are two ways to toggle this:

From the list:

1. Hover over the document row.
2. Click the globe icon in the actions column.
3. The status flips — you'll see "Document set as global" or "Document is now per-agent only."

From the detail panel:

1. Open the document and go to the Agents tab.
2. In the Availability card, click Make Global (or Global, if it's already global, to turn it off).

---

Assigning documents to agents

If a document is per-agent, you decide exactly which agents get it. There are two places to do this.

Option A: From the document's Agents tab

1. Open the document and go to the Agents tab.
2. Make sure the document is not global (global documents go to everyone, so per-agent assignment is hidden).
3. Review the Assigned Agents list — these are the agents that currently have the document.
4. Click an assigned agent's name to jump straight to that agent.
5. Click Manage Agent Assignments to add or remove agents.

The Manage Agent Assignments dialog

This dialog lists your active agents. For each one you can Assign or Remove the document with a single click.

1. Use the search box at the top to find an agent by name.
2. Click Assign next to an agent to give it the document. The row turns green and shows an Assigned badge.
3. Click Remove to take the document away from that agent.
4. Each change saves immediately (you'll see a toast like "Assigned to "Front Desk"").
5. Click Done when you're finished.

Option B: From the agent editor

You can also attach documents while you're building an agent — see the next section.

---

Attaching documents from the agent editor

When you open an agent and go to its Knowledge section, you'll find a list of all the documents in your Knowledge Base. This is the agent-side view of the same library.

The Knowledge section is titled "Knowledge" with the subtitle "Attach documents your agent can retrieve during conversations."

1. Open the agent you want to edit.
2. Go to the Knowledge section (the open-book icon in the agent's section list).
3. You'll see All Documents with a count of how many are currently attached.
4. Use the search box to find a document by name.
5. Flip the switch next to a document to attach or detach it for this agent.
6. Attached documents also appear as removable pills at the top of the section.

Each document row shows its size, its processing status (e.g. completed, failed, pending), and a global badge if it's a global document.

Important: You can only switch on documents that are completed (Ready). Documents still pending, processing, or failed are shown but cannot be toggled until they finish.

Two paths, same result. Whether you attach a document from its Agents tab in the Knowledge Base or from the agent's Knowledge section, the document ends up available to that agent. Use whichever is more convenient — the Knowledge Base view is best when working document-by-document; the agent editor is best when setting up one agent's whole knowledge set.

---

How an agent uses the Knowledge Base during a call

Once a document is Ready and available to an agent (either global, or assigned/attached to that agent), the agent can use it live during conversations.

Here's what happens behind the scenes:

1. A caller asks a question — for example, "How long do I have to return something?"
2. The agent searches its knowledge. Because documents are split into chunks and indexed for vector search, the agent finds the passages most relevant to the question — even if the caller's wording doesn't match the document exactly.
3. The agent grounds its answer in those passages, so the response reflects what's actually written in your documents rather than a guess.
4. The caller hears an accurate, on-brand answer — for example, your real refund window straight from your policy file.

This retrieval is what "grounded responses" means: the agent's answers are anchored to your real content. The more relevant and well-organised your documents are, the better the agent performs.

Availability recap: An agent can only use a document if (a) the document is Ready, and (b) it's global or assigned to that agent. A document that's still processing, has failed, or isn't assigned won't be searched.

---

Retrying and fixing stuck or failed documents

Occasionally a document fails to process, or gets stuck. The Knowledge Base gives you clear ways to recover.

Failed documents

A Failed document shows a red status pill. To fix it:

• From the list: hover over the row and click the retry (↻) icon.
• From the detail panel: open the document, go to the Details tab, and click Retry processing → in the red error panel.

The document is queued for reprocessing and returns to Pending/Processing. You'll see "Document queued for reprocessing." The Details tab also shows the specific processing error, which can help you understand what went wrong (for example, an unreadable or corrupted file).

Stuck documents

If a document seems stuck in Processing for an unusually long time, you can force it to restart:

• From the list: hover over the row and click the blue retry icon (its tooltip reads "Force retry (stuck)").
• From the detail panel: open the document and click Force retry → in the blue panel on the Details tab.

This resets the document and requeues it from scratch.

---

Deleting a document

Deleting removes a document from your Knowledge Base permanently and detaches it from any agents that were using it.

You can delete a document from three places:

• From the list: hover over the row and click the trash icon.
• From the detail panel header: click the trash icon at the top of the panel.

After deletion you'll see "Document deleted," the row disappears, and the detail panel closes if that document was open.

Deletion is immediate and permanent. If agents were relying on the document, their answers will no longer be grounded in it. Make sure you have the source file elsewhere if you might need it again.

---

Best practices for great Knowledge Base content

The quality of your agents' answers depends directly on the quality of your documents. Follow these guidelines.

Structure and formatting

• Keep each document focused. One topic per document (e.g. "Returns Policy", "Opening Hours", "Product Catalogue") works far better than one giant mixed file.
• Use clear headings and short paragraphs. Well-structured text chunks cleanly, so the agent can retrieve a precise passage.
• Prefer plain text and Markdown for facts you want quoted exactly. They're clean, editable in place, and predictable.
• Use CSV for tabular data like price lists or product specs.

Content quality

• Write in plain, direct language — the way you'd want the agent to answer. The agent grounds its wording in your text.
• State facts explicitly. Don't assume context. Instead of "the usual window," write "a 30-day return window."
• Include the questions customers actually ask. A short FAQ-style document (question, then answer) is excellent KB material.
• Avoid contradictions. If two documents disagree (e.g. different prices), the agent may surface either one. Keep a single source of truth per fact.

Maintenance

• Keep it current. Out-of-date documents lead to wrong answers. For text files, use Edit → Save & Reprocess to update in seconds.
• Remove stale documents. Delete superseded versions so the agent never retrieves old information.
• Check the status. Only Ready documents are used — after uploading or editing, confirm the document returns to Ready.
• Use global sparingly and deliberately. Make truly company-wide facts global; keep niche or campaign-specific content per-agent so agents aren't distracted by irrelevant material.

Images and scanned documents

• Images (PNG/JPG/WEBP) are best for diagrams or visual references. For text you want the agent to read reliably, type or paste it into a text/Markdown document rather than uploading a screenshot.

---

Common workflows

"Set up shared company knowledge for all my agents"

1. Upload your core documents (hours, contact info, general policy).
2. Wait for each to reach Ready.
3. Click the globe icon on each to make it Global.
4. Done — every agent now references them automatically.

"Give one agent its own specialist knowledge"

1. Upload the agent-specific document(s).
2. Wait for Ready.
3. Open the document → Agents tab → Manage Agent Assignments.
4. Assign the document to just that agent (leave it per-agent, not global).

"Update a price list the agent quotes"

1. Open the Knowledge Base and click the document.
2. On the Content tab, click Edit (text/CSV/Markdown only).
3. Change the prices and click Save & Reprocess.
4. Wait for the status to return to Ready — the agent now quotes the new prices.

"Fix a document that won't process"

1. Find the Failed document in the list.
2. Open it → Details tab → read the processing error.
3. Click Retry processing →. If it keeps failing, re-create the file as clean text and upload a fresh copy, then delete the broken one.

---

Troubleshooting

---

Quick reference

Statuses

Where to do each task

Supported file types: .pdf, .docx, .txt, .md, .csv, .json, .png, .jpg, .jpeg, .webp

The Knowledge Base is one of the highest-impact tools in the platform: a few well-written, well-organised documents can transform your agents from "sounds helpful" into "always accurate." Keep your content focused, current, and clearly assigned, and your agents will reward you with grounded, trustworthy answers on every call.

---

How retrieval works (semantic search)

When an agent needs to answer a question, it doesn't keyword-match — it runs semantic search across your uploaded content. Your PDFs, documents, and web pages are chunked and embedded, and the agent retrieves the passages most relevant to what the caller actually asked, then answers grounded in that material.

• Sources: PDFs, Word/text docs, and web page URLs (scraped and indexed).
• Per-agent assignment: restrict each agent to specific documents so it only answers from approved, relevant content — which keeps answers accurate and avoids drift.
• Mid-call: retrieval happens live during the conversation, as one of the agent's built-in tools, so the caller gets a grounded answer in the moment.