Skip to main content
By the end of this guide you’ll have a working voice agent grounded in your own content, with a workflow tool it can call mid-conversation.

​
Prerequisites

  • An Anyreach account and an active organization
  • Permission to create agents and knowledge bases in your org
  • A document (PDF, Markdown, plain text, or a URL) you want the agent to know about

​
1. Create a knowledge base

Knowledge bases let your agent answer factual questions from your content.
1

Open Knowledge Bases

From the sidebar, click Knowledge Bases, then New knowledge base.
2

Name and describe it

Give it a clear name (e.g. Product FAQ) and a one-sentence description. The description helps the agent decide when to query this KB.
3

Add a source

Click Add source. Upload a .pdf, .md, .txt, or .html file, or paste a URL to crawl. Wait for the source status to reach Ready — it goes through pending → converting → chunking → embedding → ready.
See Uploading documents for format details and limits.

​
2. Create an agent

1

Open Agents → New agent

Pick a starting template or Blank. The agent editor walks through a few sections in order.
2

Instructions

Set the opening (who speaks first and what the agent says), then write the prompt: who the agent is, how it speaks, what it tries to accomplish, and what it does at the end.
3

Voice & Speech

Pick a language and a voice.
4

Knowledge

Attach the knowledge base you just created. The agent will retrieve from it when callers ask factual questions.
5

Skip Tools & Actions for now

We’ll come back here in step 4 to expose a workflow as a tool.

​
3. Test the agent

Open the Tester panel on the agent edit page, click Start call, and talk to your agent in your browser. Iterate on the prompt and instructions until the conversation feels right. See Testing agents for the full debugging surface.

​
4. Add a workflow as a tool

Workflows let the agent take actions during a call.
1

Create a workflow

Open Workflows → New workflow. Add an Input step that accepts customer_email: string, then an HTTP step that POSTs to your CRM, then an Output step that returns { "ok": true }.
2

Publish version 1

Click Publish. Only published versions can be invoked from agents.
3

Expose it as a tool

Back on the agent edit page, go to Abilities → Add tool → Workflow. Pick the workflow, give it a name (e.g. log_lead) and a description the LLM will read to decide when to call it. Map the tool’s parameters to the workflow’s input schema.
4

Republish the agent

Click Publish on the agent. The new version is now live.

​
5. Place a real call

Assign a phone number under the agent’s Deployment section, save, and call the number.

​
Next steps

Build deeper agents

Voice tuning, guardrails, model choice, multi-tool setups.

Master workflows

Step types, expressions, triggers, and end-to-end tutorials.