Create your first AI agent

To create your first AI agent in Wassla, open the Agents page, click Hire a new agent, pick a starting template or a blank canvas, give your agent a name and role (agents are people in Wassla — Layla #Support, Omar #Onboarding, never Bot), choose Text, Voice, or Text & voice, then open the editor to write your system prompt and attach a knowledge base. The whole flow takes under five minutes on a fresh workspace.

This guide walks you through every step in order — from naming through your first test conversation.

Open the Agents page

  1. Sign in to Wassla and pick the workspace you want to work in.
  2. In the left sidebar, click Agents. You will see a grid of any agents you already have, plus a dashed Hire a new agent card at the end of the grid.
  3. Click Hire a new agent. A two-step wizard opens.

If the Hire a new agent card is greyed out, your role does not include workspace management. Ask an owner or admin on your team to either hire the agent for you or upgrade your role. See Team roles and permissions for the role matrix.

Step 1 — Choose a teammate

The first wizard step shows five starter templates and a Create from scratch tile.

  • Layla #Support — orders, returns, everyday questions
  • Tarek #Sales — qualifies leads and books demos
  • Omar #Onboarding — walks new customers through setup
  • Hana #Retention — wins back lapsed customers warmly
  • Ziad #Bookings — takes reservations and reschedules (voice)

Templates are starting points, not locked configurations. Wassla will read your company profile and knowledge base after you hire and auto-write a first-draft system prompt for the role you picked. You can rewrite every word later.

If none of the templates fit, click Create from scratch — a blank-canvas agent with a generic face you can rename and re-skin in the editor.

Click your chosen tile, then click Next.

Step 2 — Name your agent and pick the modality

Wassla treats agents as named people, not as bots. The name your customers see on every reply is the name you type here.

  1. Type a name in the Name field. Use a real first name — Layla, Omar, Maya. Avoid generic labels like "Bot", "Assistant", or "Help". The brand voice across Wassla is "Name #role", never "Bot".
  2. Pick how the agent talks to customers:
    • Text — chat-only across WhatsApp, Instagram, Facebook, SMS, and the web widget
    • Voice — phone calls via Twilio and WhatsApp Calling
    • Text & voice — both at once, with one shared persona
  3. Click Hire agent.

Voice and Text & voice agents need a paid plan. If your workspace is on the free pre-live tier, the voice options show a "Go live" badge instead of being selectable — you can still hire a text agent and test voice in the Playground for free. See Go live and start billing for the upgrade path.

What happens during the "learning" animation

After you click Hire agent, Wassla runs a short onboarding animation while it:

  1. Reads your knowledge base
  2. Learns your tone of voice
  3. Wires up your channels
  4. Runs a few test chats internally

Behind the scenes, the agent-compose edge function inspects your tenant settings and knowledge sources, then writes a first-draft system prompt, picks a default model tier, and turns on the knowledge-search tool. When the animation finishes, your agent lands in your Agents grid with a Draft status badge.

Step 3 — Open the editor and refine

Click Configure on the success screen, or click Edit on the agent card later. The editor opens with four tabs.

Personality

Set the display name, the English display name (used in agent picker dropdowns), a short description, the default conversation language, and an initial greeting that customers see when the conversation opens. The Active toggle controls whether the agent is in the rotation; leave it on.

For deeper guidance on tone and escalation, see Set your AI agent's personality.

Tools

Toggle the tools your agent is allowed to call. Today the tenant-facing tool catalogue includes:

  • Knowledge search — retrieves grounded snippets from your knowledge base before answering

More tools (booking, ticketing, payments, custom function-call tools) are exposed per-tenant as your workspace enables them. Leave Knowledge search on for every agent that needs to answer questions from your documents.

Intelligence

This is where you pick the model tier, the tone slider, the knowledge base, and the system prompt.

  1. Smarts — pick one of four tiers: Standard (fast, everyday questions), Advanced (most capable, nuanced conversations), Groq Llama (faster and cheaper Llama on Groq), or Groq Llama 8B (lightning-quick replies for short FAQ turns).
  2. Tone — drag the slider between Precise and Playful. The middle is a good default.
  3. Knowledge it can use — pick a knowledge base from the dropdown. Each agent can be scoped to a single knowledge base; build one knowledge base per audience (support, sales, billing) and assign accordingly.
  4. Instructions — the system prompt. Wassla pre-fills a first draft from your company profile and the template you picked. Edit it freely: describe what the agent should always do, what it should never do, and the topics it should escalate to a human teammate.

For prompt-writing patterns and grounding tips, see Train your AI agent.

Voice (voice-capable agents only)

Pick from three preset voices — Warm, Clear, or Friendly. Pick the spoken languages (English, Arabic, Spanish, French). Enterprise tenants on a Go-Live plan can paste a custom ElevenLabs voice id under Advanced — custom voice to use a cloned brand voice.

Attach a knowledge base

If your workspace does not have a knowledge base yet, your Knowledge dropdown shows only "None". To add one:

  1. Open Knowledge from the left sidebar.
  2. Click New knowledge base, give it a name, and add documents (PDF, Word, Markdown, plain text) or paste website URLs to crawl.
  3. Return to the agent editor's Intelligence tab and pick the new base from the dropdown.

Indexed documents are searchable within seconds; crawled URLs re-crawl every 24 hours by default. See Manage your knowledge base for formats, limits, and freshness policies.

A note on business hours

A native business-hours scheduler — set an agent to auto-pause outside work hours and route to a queue instead — is on the roadmap and not shipped yet. Today, you can pause an agent manually with the Pause button on its card, which immediately stops it from answering on every channel. Resume is one click.

Save as draft, then go live

The editor has three save actions in the footer:

  • Save draft — keeps your edits in draft_config without touching the production prompt. The Playground reads the draft, so you can test changes against real conversations without risk.
  • Promote to production — copies the draft into the live config and bumps the version number. Customers start seeing the new behaviour immediately.
  • Save — saves directly to production for a live agent (or saves the first version for a draft agent) and bumps the version.

A brand-new agent lands in Draft status. To start serving customers, click Go live on the agent card. Voice agents need a paid plan to go live; text agents go live for free on the pre-live tier.

Test before customers see it

Open the Playground from the Agents quicklinks at the top of the page. Send the kinds of questions your real customers ask. The agent shows its sources next to every answer so you can verify it is citing the right document. When an answer is wrong, click the source link, fix the document, and re-test — Wassla re-indexes automatically.