Platform Features

Knowledge & Tools

Give your agents access to your company's information and the ability to take actions in external systems — all without writing a line of code.

Knowledge Bases (RAG)

Upload your documents and the platform handles the rest: extracting text, splitting into chunks, generating embeddings, and storing them for semantic search. When an agent needs to answer a factual question, it automatically searches the relevant documents and generates a concise, accurate answer.

Supported sources

File upload: PDF, DOCX, TXT, CSV, Markdown
URL scrape: Paste any public URL — the page is fetched, cleaned, and indexed
Text paste: Paste text directly — useful for policies, FAQs, or product descriptions

How it works

After uploading, the document is processed in the background (usually a few seconds to a few minutes depending on size). Status shows as "Processing" then "Indexed" when ready.

During a call, when the agent's Knowledge Base Search tool is triggered, the platform:

Generates a semantic embedding of the caller's query.

Runs a vector similarity search over your indexed documents.

Returns the top matching passages (scoped to this agent's assigned documents only).

Asks the AI to synthesize a 1–2 sentence voice-friendly answer.

Document assignment

Documents are not shared across all agents by default. You must explicitly assign documents to each agent. This ensures Agent A can't accidentally answer questions from Agent B's knowledge base. Go to Knowledge → Agent tab or the Agent's configuration to assign documents.

Re-indexing

If a document changes, delete it and re-upload, or use the Re-index button on the document. Embeddings are regenerated automatically.

Custom Tools

Tools let agents take actions mid-conversation — not just talk. Create tools in Knowledge & Tools → Custom Tools tab.

HTTP API tool

Call any REST endpoint during a conversation. The response is available to the agent in the same turn.

json

{
  "name": "check_order_status",
  "type": "http",
  "http_method": "GET",
  "http_url": "https://api.yourstore.com/orders/{{order_id}}",
  "description": "Check order status when customer asks about their order",
  "when_to_use": "When the caller asks about delivery, tracking, or order status"
}

Use {{variable}} in the URL and body to inject values from the conversation.

Webhook tool

Fire-and-forget POST to Slack, email, or your CRM when something happens in a call. Use it for notifications or to log data.

Code tool (Python)

Write a small Python snippet that runs server-side. The arguments dict contains values from the conversation. Set the result variable to return data to the agent.

python

# Example: calculate discount
amount = float(arguments.get("amount", 0))
tier = arguments.get("customer_tier", "standard")
discount = 20 if tier == "premium" else 10
result = f"You qualify for a {discount}% discount. Final amount: ₹{amount * (1 - discount/100):.2f}"

Tool assignment

Like knowledge bases, custom tools must be assigned to specific agents. Go to Tools → Agent tab or the Agent's configuration to assign tools.

MCP Servers

MCP (Model Context Protocol) is an open standard for connecting AI agents to external data sources and tools. If you have an MCP-compatible server, you can connect it to Converse and all its tools become available to your agents automatically.

Go to Knowledge & Tools → MCP Servers tab:

Click Add MCP Server.

Enter the server URL and any authentication headers.

Click Connect — Converse discovers all available tools and caches the list.

Assign the server to agents — its tools are automatically registered.

Supported transports: HTTP (JSON-RPC 2.0) and SSE (Server-Sent Events).

Testing

Use the RAG test panel on the Knowledge page to run a search query and see what the agent would find and say. Use the test button on each custom tool to execute it with sample inputs and inspect the response.