> ## Documentation Index > Fetch the complete documentation index at: https://docs.deepblocker.ai/llms.txt > Use this file to discover all available pages before exploring further. # Create an attacker > Every field on the attacker form, from the name the caller answers to and the voice it speaks with, to the tone, the opening line and the language model. An attacker is the caller: a name, a role, a company, a voice, a tone and an opening line. What the caller says on the call comes from the scenario you pick when you build a campaign. One attacker can be used by many campaigns. > [!NOTE] > > Before you start: you need a voice. A ready-made voice works right away; a cloned voice needs consent recorded first. See [Add a voice](https://docs.deepblocker.ai/vishing/voices.md). ## Before you start * A voice, cloned or ready-made. * A clear idea of who the caller is. A real-sounding name, role and company make the call land. ## Steps **1. Click New attacker** On the Red Team tab, click **New attacker** at the right end of the segment bar. You can also click **Create an attacker for this voice** on a voice. **2. Say who they are** Fill in **Name**, **Role / title** and **Company**. Only the name is required. The form says: "This name is said on the call. The caller is told to always answer to it, even if asked directly who is calling." **3. Pick the voice** Choose **Clone or record a voice** or **Use a ready-made voice**. * **Use a ready-made voice** shows **Sarah** and **George** first, with **More voices (4)** for Aria, Roger, Daniel and Jessica. The form notes: "A ready-made voice gets you started right away. Clone a voice later for a more convincing, personalised call." * **Clone or record a voice** shows a **Cloned voice** list of the voices in your workspace that have consent. If you have none, the form offers to clone one. **4. Tune the voice (optional)** A cloned voice shows four sliders and a model. Leave them alone for a first attacker. **5. Set the tone** Under **Tone of voice**, pick up to four presets and add your own words, or leave it empty. The tone holds for the whole call. A campaign can change it for one run. **6. Write the opening line** **The opening line** is required. The form says: "The first thing the caller says. It always starts with the name above by default." Up to 2,000 characters. **7. Click Create attacker** The attacker appears in the **Attackers** segment with its voice. Next, build a campaign with it, or click **New campaign** on its row. ## Every field ### Who they are | Field | What it does | Default or limit | | - | - | - | | **Name** | Said on the call. The caller answers to it when asked who is calling | Required, 200 characters | | **Role / title** | The job the caller claims | 200 characters | | **Company** | The organisation the caller claims to be from | 200 characters | ### Voice | Field | What it does | Default or limit | | - | - | - | | **Clone or record a voice** / **Use a ready-made voice** | Which kind of voice the attacker speaks with | Ready-made when you have no cloned voice | | **Ready-made voice** | One of the six stock voices | Sarah and George shown first | | **Cloned voice** | A voice from your workspace with consent recorded or attested | Required when cloning is chosen | | **Model** | The speech model for a cloned voice | **Eleven Turbo v2.5 (recommended)** | | **Speed** | How fast the caller talks | Shown as a multiplier, for example 1.00x | | **Stability** | How steady the delivery is; lower is more varied | Shown as a number between 0 and 1 | | **Similarity** | How closely the clone sticks to the sample | Shown as a number between 0 and 1 | | **Style** | How much expression is added | Shown as a number between 0 and 1 | The models you can pick for a cloned voice: | Model | The form's note | | - | - | | **Eleven Turbo v2.5 (recommended)** | Low, consistent latency. Default for cloned voices | | **Eleven Flash v2.5** | Fastest replies. A touch less rich than Turbo | | **Eleven v4 Turbo** | Most expressive, but routed through our bridge, so latency is higher and less consistent | | **Eleven Multilingual v2** | Steadier on long sentences, but slower to start speaking | ### Tone of voice Pick up to four. Each one changes how the caller speaks for the whole call. | Preset | How the caller sounds | | - | - | | **Panicked** | Fast, short bursts, breathing audibly, repeating themselves | | **Stressed** | Tense, hurried, a little curt, checking the time | | **Calm** | Even pace, measured, patient | | **Cool and detached** | Unhurried, businesslike, few words, no warmth | | **Warm** | Friendly, on first-name terms quickly, grateful | | **Impatient** | Clipped, interrupting, pushing for it to be done now | | **Apologetic** | Sorry to bother them, softening every request | | **Confident** | Assured, direct, states what happens next | Your own words go in the box next to the presets, up to 200 characters, for example "whispering because people are around". Brackets are removed. ### The opening line | Field | What it does | Default or limit | | - | - | - | | **The opening line** | The first sentence the caller says | Required, 2,000 characters | | **Use the suggested opening line** | Fills the opening line from a saved research profile you pick. It only fills when you click the button | Shown when the workspace has research profiles | ### Advanced Click **Advanced** to show two more fields. | Field | What it does | Default | | - | - | - | | **Language** | The language the caller speaks. The list is English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, Chinese, Japanese, Korean, Hindi and Turkish | English | | **Language model** | The brain behind the caller's replies, grouped by provider. Each option's label gives a rough speed | **Claude Haiku 4.5 (fast and smart, recommended)** | The language model groups are **ElevenLabs Realtime (ElevenLabs runs the brain, fastest)**, **Fast models you pick (ElevenLabs hosted)**, **Anthropic**, **OpenAI** and **Google**. Faster models reply sooner; smarter models handle an awkward conversation better. ## Editing and deleting * **Edit** on an attacker's row opens the same form. The page says: "The change applies to every campaign that uses this attacker." * The delete button on the row asks you to confirm. Finished campaigns keep their results. A draft that uses the attacker must pick another one. * An attacker cannot be deleted while a campaign using it is scheduled, running or paused. The message reads "A campaign still uses this attacker and is scheduled, running or paused. Stop or finish it first." ## What happens next Build a campaign and pick this attacker on the **Who calls** step. See [Build a campaign](https://docs.deepblocker.ai/vishing/campaigns/create.md). Then hear it: [Take a test call in your browser](https://docs.deepblocker.ai/vishing/campaigns/test-call.md). ## Problems > [accordion] The Cloned voice list is empty > > Only voices with consent appear. Open [Voices](https://docs.deepblocker.ai/vishing/voices.md) and check each voice's status. Or switch to **Use a ready-made voice** to get going. > [accordion] The form says the voice was deleted or is no longer available > > The cloned voice this attacker used is gone. Pick another voice under **Cloned voice** and click **Save attacker**, otherwise this attacker's campaigns cannot call. > [accordion] Create attacker stays greyed out > > **Name** and **The opening line** are both required, and a cloned voice must be chosen when you picked **Clone or record a voice**. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.