Configuring Your Agent

Start With A Simple Editing Strategy

Before you change any fields, decide what kind of agent you are shaping. A public-facing sales agent needs a different level of polish and restraint than a private study partner. A family assistant needs a different tone than a call-center receptionist. That clarity will make the rest of the editor much easier to use.

For most first-time users, the safest order is:

1. AI Models
2. Voice and Realtime
3. Essentials and identity
4. Soul and prompts
5. Memory and knowledge
6. Security
7. Heartbeat, schedules, skills, and triggers

That order mirrors the real logic of the product. First decide how the agent thinks and speaks. Then define who it is. Then teach it. Then control it. Then automate it.

AI Models: Choose The Brain First

The AI Models section controls the reasoning core of your agent. This is where you choose the provider, the specific model, the reasoning level, and the fallback chain. If you have not already read the separate guide on choosing your AI brain, do that after this section if anything feels unclear.

The Provider field chooses the AI company powering your agent’s main reasoning. In the current editor, the primary choices include OpenAI, Anthropic, Google, and xAI, along with broader advanced-provider surfaces like OpenRouter and other model networks. For most users, focus on the four frontier providers first. Choose OpenAI for general-purpose reliability and strong voice compatibility. Choose Claude when you want deeper analysis and more careful reasoning. Choose Gemini when you want multilingual and cost-aware performance. Choose Grok when personality and live social context matter.

The Model field chooses the specific version inside that provider. Do not treat this like a permanent marriage. It is a working selection. If your agent feels too shallow, you can raise the tier or change the model. If it feels too expensive for routine work, you can lower the tier or use a lighter model.

The Reasoning Level cards are the easiest way to think about this section. Quick is for fast, low-cost work. Standard is the best default for most agents. Deep Think is for more difficult tasks. Maximum routes the request into SuperIntelligence, where NetShow asks all four frontier models and synthesizes one best answer. If you do not know what to choose, use Standard. If the answer really matters, use Maximum.

The Fallback fields matter whenever uptime matters. These tell NetShow what to try if the primary provider is unavailable. A good business default is to keep OpenAI as a backup because it is broad and dependable. If you run a customer-facing agent or a phone workflow, fallback settings are worth taking seriously.

You may also see API-key-related fields in this area, especially for OpenRouter or externally supplied model access. Only fill those in if you are intentionally using your own provider account. If you are not sure, leave those fields alone and use the platform defaults until you have a reason to change them.

Voice And Realtime: Decide How The Agent Speaks

The Voice and Realtime section controls live spoken conversations. This is separate from the text chat brain above. That distinction matters. Your main reasoning model can be one thing, while your voice delivery stack can be another.

The first important choice here is Voice Provider. NetShow lets you choose from OpenAI Realtime, Gemini Live, Grok Voice, Hybrid, and ElevenLabs-related setups. In plain language, OpenAI is usually the safest choice if you want the most proven spoken experience. Gemini Live is strong when language flexibility and more adaptive conversation matter. Grok Voice is for a more expressive, personality-rich feel. Hybrid mode is especially useful when you want one provider to think and OpenAI to handle the spoken delivery.

The Realtime Model field controls the live conversation model for voice interactions. This is not necessarily the same as your main text model. Most users should leave this on the recommended realtime option unless they know they need a different one.

The Realtime Voice field is where you choose the actual speaking voice. The current editor includes a wide range of labeled voice styles, including warmer, more professional, more gentle, and more expressive options. Pick the one that matches the role of the agent. A business phone assistant should sound steady and credible. A family helper can feel more approachable. A coaching voice can be more energetic.

If you select Gemini Live, you may also see language, affective dialog, and camera input options. Use the language field when you want one default spoken language. Turn on affective dialog only if you want the voice system to adapt more noticeably to emotional cues. Camera input is useful only for setups that genuinely need visual conversation.

If you select Grok Voice, you can enable expressive mode. This makes sense for creator-style experiences, social personalities, or public-facing agents that benefit from more character. It is usually unnecessary for formal business roles.

The Turn Detection fields control how the agent decides you are finished speaking. If the agent interrupts too quickly, increase silence duration or reduce sensitivity. If it waits too long and conversations feel awkward, make it a little more responsive. The defaults are usually fine at first.

The Audio Settings fields manage noise reduction, noise mode, and truncation behavior for long conversations. Unless you are diagnosing a specific voice problem, the defaults here are usually a good place to stay.

The Tool Namespace field decides what the voice experience can access. Keep this limited to what the agent genuinely needs. More tools do not automatically make better conversations.

If you want the agent to handle phone calls, the Phone / SIP settings let you enable phone behavior, assign a number, and configure sideband communication. Do not turn this on just because it sounds advanced. Turn it on when the agent genuinely needs to answer or manage live calls.

ACT Settings: Shape The Communication Experience

ACT stands for Agent Communication Terminal. This is the workspace that controls how the agent behaves in richer conversation surfaces, especially where there is more than a simple text exchange happening.

ACT settings are the place to pay attention to wake words, sleep words, introductory visibility, and other behavior that shapes how the agent enters and exits an interaction. If your agent should feel like a lively companion or a voice-first helper, these fields matter. If the agent is mostly a quiet utility tool, you can keep them simpler.

The key question in ACT is not “What is technically possible?” It is “How much ceremony should this conversation have?” A wake word, an intro message, and visible cues make sense for a voice companion. They may be unnecessary for a straightforward website helper. Match the amount of ACT behavior to the kind of experience you want.

Quick Replies And Agent Questions: Give People An Easy Starting Point

The editor includes an Agent Questions or pre-set prompt area that acts like quick replies. Think of this as the first rung of the ladder for the user. These are the questions your agent can ask or answer right away, or the starter prompts you want people to click instead of staring at a blank input.

Use this section to seed useful first interactions. For a sales agent, that might be “What package is right for me?” or “Can I book a demo?” For a family assistant, it might be “What is on my calendar today?” For a learning partner, “Quiz me on chapter three.” Keep these short and obvious. Their job is not to impress. Their job is to help people get moving.

If you add too many, they stop helping. Three to six strong starters is usually enough.

Heartbeat: Make The Agent Proactive

Heartbeat is one of NetShow’s more distinctive features because it moves the agent from reactive to proactive. When Heartbeat is enabled, the agent can check in on a schedule instead of waiting to be asked something first.

The Enable Heartbeat toggle turns that behavior on or off. The Check-in Frequency field lets you choose daily, weekly, or custom scheduling. The Heartbeat Tasks field is where you tell the agent what to watch for or follow up on.

This field works best when the tasks are concrete. “Check my life” is too vague. “Review my calendar for tomorrow, look for conflicts, and remind me about anything starting before 9 AM” is good. “Check for unanswered sales leads and remind me if a prospect has gone quiet for three days” is good. Heartbeat works when it has a real job.

If you are not ready for proactive behavior, leave this off at first. It is a high-value feature, but it is better when you know what kind of follow-up you want.

Essentials And Identity: Tell NetShow Who This Agent Is

The Essentials area is where you edit the agent’s basic identity. This includes fields like Agent Name, Agent Title, Agent Tags, Agent Description, and in many cases contact fields such as phone number and email. This is also where introductory media and intro text live.

Use the Agent Name for the user-facing identity. Use the Title to explain the role more clearly. The Tags field is useful for describing what the agent is about, how it should be found, and what themes define it. The Description should read like an internal one-paragraph role brief.

The Agent Phone Number and Agent Email fields matter when the agent is tied to real customer communication or workflows. If the agent is strictly internal or experimental, you may not need these right away.

The Introductory Text field is important. This is often the first real sentence the user sees from the agent. It should sound natural, useful, and specific. Avoid generic phrases like “Hello, how may I assist you today?” unless the context truly demands it. A good intro tells the user what kind of help is available immediately.

Appearance: Avatar, Intro Media, And Presence

The appearance-related sections control what the agent looks like and, in richer surfaces, how it appears through idle video or avatar systems. In the current editor this includes avatar upload, current image selection, idle video generation, and additional avatar-provider features such as Simli or Hedra-related surfaces.

Most users only need to make one good visual choice here. Pick an avatar image that is clear, stable, and appropriate for the job. If you use intro video or idle video, keep it calm and professional unless the use case is intentionally more expressive.

Do not let the visual side slow you down. A decent avatar is enough to launch. Come back later if embodiment becomes an important part of the experience.

Soul: Define Personality, Values, And Boundaries

The Soul section is one of the most important parts of the editor because it tells the agent how to feel, not just what to do. The main fields here are Personality & Vibe, Core Values, Boundaries, Communication Style, and Emoji Style.

The Personality & Vibe field is where you describe how the agent should come across. Write this as if you were briefing an actor or a new hire. “Warm, calm, clear, and non-judgmental” is a strong description. “Funny and cool” is weak. Be specific enough that the agent can actually follow it.

The Core Values field tells the agent what principles should guide it. This is useful for customer service tone, brand standards, or personal assistant behavior. Values like honesty, clarity, patience, privacy, and follow-through can all belong here.

The Boundaries field is crucial. This is where you explain what the agent should never do, overpromise, speculate about, or handle without escalation. If your agent is public-facing, do not leave this vague.

The style and emoji fields are finishing tools. Use them to reinforce tone, not replace it.

Memory: Control What The Agent Remembers

Memory is where the agent becomes more than a one-message responder. In the editor, this includes Auto-Learn, Auto-Archive, Privacy Mode, Memory Retention, Max Memory Size, and manual memory entry.

Turn on Auto-Learn if you want the agent to capture useful facts from conversations. Turn on Auto-Archive if you want older memories compressed instead of lost. Keep Privacy Mode on unless you have a very unusual reason not to. That setting helps prevent the agent from storing highly sensitive information inappropriately.

The Retention and Max Memory Size fields are really about cost and usefulness. More memory can make the agent more helpful, but it can also increase token use. For most users, the recommended middle settings are the right place to start.

The manual memory field is useful for facts the agent should know immediately. Add things like user preferences, brand standards, recurring routines, or business rules. Keep each memory fact clear and self-contained.

Instructions: Write The Main Prompt Carefully

The Agent Prompt field is the core instruction set for the agent. If the Soul section defines the personality, this section defines the operating logic. This is where you explain the job, the workflow, the tone, the priorities, what to ask, when to escalate, and how the agent should make decisions.

This field does not need to sound technical. In fact, plain language is usually better. Write it as if you were training a human assistant. Tell the agent who it helps, what it should ask first, what information it should collect, what kind of answer style to use, and when it should stop and hand off.

A bad prompt is vague and dramatic. A good prompt is specific and practical. After you write it, read it once as if you were the agent. If there is ambiguity, tighten it.

Demographics: Use This Only When It Helps

The Demographics section is more specialized. It includes fields around gender, age, birth-date range, personalities, philosophical ideology, state or demographic background, and additional demographic detail. This part exists because some agents are meant to embody a more specific character, tone, or perspective.

Do not feel obligated to fill all of this out. Use demographics only when they genuinely improve the experience. A memorial agent, roleplay-style guide, or character-based educational agent may benefit from more detail here. A straightforward business support agent usually does not need much of it.

If you do use this section, make sure it supports the job instead of distracting from it.

MD Files: The Agent’s Core Configuration Files

The markdown configuration files are one of NetShow’s most powerful advanced features. In the editor, you can work with files like SOUL.md, MEMORY.md, KNOWLEDGE.md, HEARTBEAT.md, and USER.md. Think of these as the agent’s living operating manual.

SOUL.md is where you define personality, communication style, values, and boundaries in a durable format. MEMORY.md is where persistent memory logic and higher-level memory behavior live. KNOWLEDGE.md can hold durable reference content that should shape the agent’s understanding. HEARTBEAT.md supports proactive behavior and check-in logic. USER.md is useful for persistent user-specific guidance.

If the regular form fields feel like dials and switches, the markdown files feel like authoring the deeper source material behind the agent. You do not need to master them immediately, but they are worth learning because they make the agent more coherent over time.

Start by using them for things you want to express in normal language, not for clever formatting. Clear markdown beats complicated markdown.

API Keys And Advanced Model Access

Some agents need outside keys, especially when using advanced routing or provider-specific options such as OpenRouter. If you see fields for API keys, only fill them in when you intentionally want the agent to use your external account or a specific provider pathway.

If you are not sure why a field is there, leave it empty. The editor contains both beginner and advanced surfaces. Empty advanced fields are often safer than guessed values.

Fine-Tuning: Advanced And Optional

NetShow includes fine-tuning or dataset-oriented surfaces in the broader editor area, but this is not something most users need to touch on their first pass. Think of fine-tuning as a specialist tool for teams that have enough example data and enough clarity to train or structure advanced behavior deliberately.

If your agent is not yet performing well, do not jump to fine-tuning first. Fix the prompt, the soul, the knowledge, and the memory before you go there. Those improvements usually deliver more value, faster.

Security: Decide What The Agent Is Allowed To Do

The Security section controls the guardrails around the agent. In the current editor, this includes protections such as prompt injection guard, action approval, audit logging, sandboxing, sensitivity controls, session action limits, and retention settings.

Leave the core protections on unless you have a tested reason to change them. Prompt injection guard is especially important for public or semi-public agents. Action approval matters whenever the agent can send messages, call APIs, delete data, or trigger meaningful workflows.

The sensitivity controls are where you decide which actions require explicit approval. For many users, email sending, payment actions, data deletion, and external API calls should all stay gated. It is better to add freedom slowly than to remove risk after something goes wrong.

For most agents, standard action limits and a reasonable retention period are enough. If the agent is handling sensitive or regulated interactions, be more conservative.

Knowledge, Skills, Schedules, And Triggers

The newer component-based panels handle some of the most practical configuration work in the editor.

The Knowledge section is where you upload files, add URLs, and manage the material your agent uses as reference. Use this when the agent needs to know your products, policies, processes, or documents.

The Skills section is where you expand what the agent can do and, in some cases, capture or attach skill-like capabilities more deliberately. Use this after the identity and prompt are stable.

The Scheduler section lets you build recurring actions such as check-ins, follow-ups, reminders, or reports. This is especially useful when paired with Heartbeat.

The Triggers or Agent Action Links section lets you connect keywords to actions like opening an iframe or executing a workflow. This is powerful, but it should be used carefully. Add triggers only for patterns you genuinely want to automate, and test them before relying on them in live customer interactions.

A Good First Configuration Pass

If you want a practical first pass through the editor, do this. Confirm the AI provider and set Standard reasoning. Choose the voice provider and voice. Update the name, title, description, and intro text. Write the soul section in clear language. Tighten the main prompt. Upload two or three high-value knowledge sources. Turn on privacy-friendly memory. Review the security defaults and keep them conservative. Then test the agent with real conversations before touching advanced options.

That sequence gives you an agent that feels coherent, useful, and reasonably safe without forcing you to master every advanced feature in one session.

Final Advice: Edit For The Next Conversation, Not The Perfect Future

The best way to use the editor is to improve the next conversation, not to design a perfect forever configuration in one sitting. Every field should answer a practical question: will this make the agent more helpful, more trustworthy, or more aligned with the job it is supposed to do?

If the answer is yes, make the change. If the answer is “maybe someday,” leave it for later. NetShow is powerful because you can keep refining. You do not need to do it all at once.