Skip to main content

​
Knowledge Base

The Knowledge Base allows you to provide your agents with domain-specific information. During a conversation, the agent automatically retrieves the most relevant content to answer caller questions accurately — instead of relying on the LLM’s general training data.

​
How It Works (RAG)

Callem Studio uses Retrieval-Augmented Generation (RAG) to give your agent real-time access to your data:
1

Content ingestion

When you create a knowledge item, the text is split into chunks and each chunk is converted into a vector embedding.
2

Vector storage

Embeddings are stored in a vector database for fast similarity matching.
3

Real-time retrieval

During a call, the caller’s question is also converted into a vector. The system finds the most similar chunks from your knowledge base.
4

Context injection

The retrieved chunks are injected into the LLM’s context alongside the system prompt, giving the agent accurate, up-to-date information to answer with.

​
Creating a Knowledge Item

  1. Navigate to Build > Knowledge Base
  2. Click Add
  3. Fill in:
    • Name: a descriptive label (e.g. “Product Catalog Q3 2025”, “FAQ - Returns Policy”)
    • Description: optional summary explaining what the content covers
    • Content: the actual text your agent should know

​
Linking to an Agent

Once created, you need to link the knowledge base to your agent:
  1. Go to Build > Agents and select your agent
  2. Open the Model tab → Knowledge & Tools section
  3. Select one or more knowledge bases from the dropdown
  4. (Optional) Adjust Retrieval Settings

​
Retrieval Settings

Fine-tune how the agent retrieves content:
SettingDescriptionDefaultGuidance
Max documentsMaximum text chunks retrieved per query (1-10)6More chunks = more context but higher token usage. Start with 4-6 and increase if the agent frequently lacks information.
Relevance thresholdMinimum similarity score (0.0-1.0) for a chunk to be included0.6Higher = stricter matching, fewer but more relevant results. Lower = more results but may include loosely related content.
If the agent frequently answers “I don’t have that information” when the data exists in your knowledge base, try lowering the relevance threshold or increasing max documents.

​
Content Guidelines

Write content as you’d want the agent to say it. Avoid internal jargon unless the agent should use it with callers. The LLM will rephrase as needed, but clear source text gives better results.
Use ## Heading to separate topics. Each heading creates a natural chunk boundary, improving retrieval accuracy. Don’t put unrelated information under the same heading.
One knowledge item per topic or document. A “Business Hours” item and a “Pricing” item will retrieve better than a single “Everything About Our Company” item.
Instead of “Our service is affordable”, write “Our Standard plan costs €49/month, Professional is €99/month.” The agent can only relay information it has.
Adding common questions and their answers helps the retrieval engine match caller questions to relevant content. Example: “Q: What are your hours? A: We are open Monday to Friday, 9am to 6pm.”
Outdated knowledge leads to wrong answers and erodes caller trust. Delete and recreate items when the underlying information changes — pricing, availability, policies, etc.

​
Example Content Structure

## Business Hours
We are open Monday to Friday, 9am to 6pm.
Saturday from 10am to 2pm. Closed on Sundays and public holidays.
Holiday closures: December 25, January 1, May 1.

## Return Policy
Items can be returned within 30 days of purchase with original receipt.
Refunds are processed within 5-7 business days to the original payment method.
Sale items are final sale and cannot be returned.
Exchange is available in-store within 14 days.

## Pricing
- Standard plan: €49/month — up to 500 calls/month, 1 agent
- Professional plan: €99/month — up to 2,000 calls/month, 5 agents
- Enterprise: contact sales at sales@acme.com for custom pricing
All plans include a 14-day free trial with no credit card required.

## Frequently Asked Questions
Q: Can I upgrade my plan mid-month?
A: Yes, upgrades are prorated. You only pay the difference for the remaining days.

Q: Do you offer annual billing?
A: Yes, annual billing gives you 2 months free (pay for 10, get 12).

​
Multiple Knowledge Bases

You can link multiple knowledge bases to a single agent. The retrieval system searches across all linked knowledge bases simultaneously and returns the most relevant chunks regardless of which knowledge base they belong to. Common patterns:
  • Product catalog + FAQ + Policy documents — different types of content, all accessible
  • Language-specific — separate knowledge bases per language
  • Temporal — archive old content in a separate KB, keep current content in the primary one

​
API Access

Knowledge base items can be managed programmatically via the Public API:
  • GET /v1/knowledge — list all items
  • GET /v1/knowledge/:id — get a specific item
  • POST /v1/knowledge — create a new item (automatically vectorized)
  • DELETE /v1/knowledge/:id — delete an item
This is useful for integrating with CMS systems, automated content pipelines, or keeping your knowledge base in sync with external data sources.