Workspaces, Profiles, Agents, Skills, and Voice Intelligence

Vocoding is built around four main configuration layers: Workspaces, Optimization Profiles, Agents, and Skills. Voice Intelligence lives inside profiles as the personalization layer for your own words, shortcuts, style, and protected rules.

The short version:

Workspaces define where you are working, Profiles define what kind of output you want, Agents define who should reason about the task, and Skills define reusable knowledge the agent can apply.

---

Configuration Layers

Layer	User meaning	What it controls	Example
Workspace	The project or context you are working in	Which resources are available and what background context applies	"Vocoding launch", "Client CRM project", "Personal writing workspace"
Optimization Profile	How the text should be transformed	Output intent, writing structure, cleanup level, dictionary/snippet behavior, final format	Email Draft, Personal Voice Compiler, Prompt Optimizer
Agent	Who should handle the task	Specialist behavior, domain expertise, routing for task type	Frontend Architect, LinkedIn Strategist, QA Reviewer
Skill	Reusable knowledge or workflow	Specific instructions, playbooks, domain rules, checklists	Accessibility audit, profile creator, enterprise review

Provider and model settings are separate. They choose the engine that runs the request, such as local Ollama, Groq, OpenRouter, or another provider. Profiles, agents, skills, and workspaces choose the behavior.

---

Workspaces

Use a Workspace when the same project, client, or team context should apply across many tasks.

Workspaces can contain:

• Project context, stack, constraints, links, and rules.
• Assigned agents, profiles, and skills.
• Curated or global resource modes.
• Team vocabulary and repeated project assumptions.

Examples:

• "Vocoding Product Launch"
• "Client A Website"
• "Personal Writing"
• "Legal Drafting"
• "React SaaS Project"

---

Optimization Profiles

Optimization Profiles are the behavior and output layer. They control how Vocoding turns raw speech or typed input into final text.

A profile is not an agent. It does not represent a specialist person. It is an output contract.

Use a profile to:

• Clean up dictated text.
• Draft an email.
• Format meeting notes.
• Convert a rough idea into a prompt for another AI tool.
• Preserve exact technical terms and expand personal snippets.

Output Intents

Output intent	Meaning	Use when	Example input	Example output
Text Transformer	Preserve the user's text and improve it	The user already has the message, but wants it corrected or polished	"Hola Pedro confirmo la reunion manana"	"Hola Pedro, confirmo la reunión de mañana."
Artifact	Create a final deliverable	The user wants a finished email, post, report, proposal, reply, or note	"Escribe a Pedro para confirmar la reunión de mañana"	A full email with greeting, body, and closing
Prompt Compiler	Create instructions for another AI or tool	The output is a prompt to paste into a coding assistant, chat model, or image model	"Hazme una app de tareas"	A structured implementation prompt with requirements and constraints

Text Transformer changes the text itself. Artifact creates the final content. Prompt Compiler creates instructions for another system.

Text Transformer

Text Transformer profiles are for corrected or preserved text. They are ideal for dictation cleanup, grammar correction, personal dictionaries, snippets, and technical term correction.

Important rule: a Text Transformer profile must not convert ordinary dictated text into a task or prompt.

If you say:

"Estamos probando Acme Analytic y North Star API."

The profile should return:

"Estamos probando Acme Analytics y Northstar API."

It should not turn the sentence into an instruction like "Verify the processing of Acme Analytic and North Star API."

Artifact

Artifact profiles create finished pieces of content.

Good for:

• Emails.
• LinkedIn posts.
• Customer replies.
• Investor updates.
• Meeting notes.
• PR descriptions.
• Product listings.
• Legal or clinical drafts.

Example input:

"Escribe a Pedro para confirmar que viene mañana a la reunión por Meet a las 5."

Example output:

"Hola Pedro,

Te escribo para confirmar tu asistencia a la reunión de mañana por Meet a las 17:00.

Gracias, Carlos"

Prompt Compiler

Prompt Compiler profiles create prompts for another AI or tool.

Good for:

• Codex tasks.
• AI coding assistant tasks.
• Cursor prompts.
• Image prompts.
• JSON or XML prompts.
• Deep research prompts.
• Structured implementation requests.

Example input:

"Crea una pantalla para gestionar clientes con filtros y tabla."

Example output:

"Build a customer management screen with a searchable table, filters, pagination, empty state, loading state, and responsive layout. Use the existing design system..."

The output is not the final app or final email. It is the instruction for another model or tool.

---

Personal Voice Compiler

Personal Voice Compiler is a Text Transformer profile focused on voice dictation.

It should:

• Preserve the original sentence and speaker intent.
• Correct technical phrases.
• Expand personal snippets.
• Respect compound phrases, not only single words.
• Keep token cost low by using compact rules.

Spoken or mistranscribed phrase	Correct output
Acme Analytic	Acme Analytics
North Star API	Northstar API
Project Blue	ProjectBlue
Data Lake Pro	DataLake Pro
mi email	user-configured email
mi LinkedIn	user-configured LinkedIn URL

Dictionary entries and snippets should support phrases with spaces. Users often need corrections like Acme Analytic to Acme Analytics, not just single-word replacements.

---

Voice Intelligence

Voice Intelligence teaches Vocoding how you speak: your names, terms, shortcuts, snippets, style, and protected rules.

Dictionary

Use Dictionary for words or phrases that must be corrected or preserved.

Examples:

• Acme Analytic -> Acme Analytics
• North Star API -> Northstar API
• V coding -> Vocoding
• O Lama -> Ollama

Dictionary is best for names, brands, product names, technical terms, acronyms, and common transcription mistakes.

Snippets

Use Snippets for shortcuts that expand into longer text.

Examples:

• mi email -> your email address.
• mi LinkedIn -> your LinkedIn profile URL.
• intro email -> a reusable email opening.
• demo link -> a booking or demo URL.

Feature	Dictionary	Snippet
Purpose	Correct a term	Expand a shortcut
Typical size	Short word or phrase	Longer text
Example trigger	"Acme Analytic"	"mi firma"
Example output	"Acme Analytics"	Full email signature

Styles

Styles control tone and polish. Examples include direct and professional, friendly and concise, technical and precise, sales-oriented, and founder voice.

Styles should obey hard rules and dictionary entries.

Hard Rules

Hard Rules are instructions the system must not break.

Examples:

• Never change protected brand names.
• Never convert dictated text into an action prompt in Personal Voice Compiler.
• Preserve code identifiers in developer mode.
• Keep Spanish/English mixed text as spoken unless translation is requested.

---

Agents

Agents are specialists. They are useful when the task needs domain judgment, not just output formatting.

Examples:

• Frontend Architect.
• Backend API Designer.
• Marketing Strategist.
• QA Reviewer.
• Security Auditor.
• LinkedIn Content Specialist.

Use an Agent when the task needs expertise:

"Analyze this onboarding flow and propose UX improvements."

Profiles and agents can work together:

Resource	Selection
Workspace	Vocoding Launch
Agent	LinkedIn Strategist
Profile	LinkedIn Post
Skill	Brand voice checklist

The workspace supplies product context. The agent supplies strategy. The profile controls the final format. The skill supplies reusable rules.

---

Skills

Skills are reusable playbooks or knowledge packs. They are not profiles and not agents.

Use Skills for:

• A checklist.
• A workflow.
• A domain method.
• A repeatable process.
• A validator or review protocol.

Examples:

• Enterprise Review.
• Accessibility Audit.
• Profile Creator.
• Security Review.
• Spec Kit.

Agents decide how to approach the task. Skills give them specialized methods.

---

Decision Tree

1. Do you want to organize a project or client context? Use a Workspace.

2. Do you want to change the shape of the output? Use an Optimization Profile.

3. Do you want the system to behave like a specialist? Use an Agent.

4. Do you want reusable rules, methods, or checklists? Use a Skill.

5. Do you want to teach Vocoding personal words, phrases, shortcuts, or style? Use Voice Intelligence inside a profile.

6. Do you want better, faster, cheaper, or private model execution? Configure Provider and Model settings.

---

Workflow Examples

Workflow	Configuration used	Result
Personal dictation	Workspace: Personal Context; Profile: Personal Voice Compiler; Voice Intelligence Dictionary enabled	Corrected text that preserves the original sentence
Email drafting	Profile: Email Draft; Output intent: Artifact	Full email ready to send
AI coding prompt	Profile: Prompt Optimizer Dev; Output intent: Prompt Compiler	Structured prompt with goal, context, constraints, acceptance criteria, and tests
Expert review	Agent: Security or Architecture Reviewer; Skill: Enterprise Review or Security Checklist; Profile: analysis/report output	Expert analysis with risks and recommendations

---

Common Mistakes

Using an Agent when a Profile is enough

If you only want an email format, use a profile. Do not create an agent for formatting alone.

Using a Prompt Compiler for text cleanup

Prompt Compiler is for creating prompts for another AI or tool. It should not be used for ordinary dictation cleanup.

Putting personal data in system profiles

System profiles are global. Personal data such as email, LinkedIn, address, or private snippets should live in custom profiles, workspace context, or user-level Voice Intelligence.

Putting dictionaries only inside a prompt

Dictionary and snippet rules should be structured when possible so Vocoding can validate and apply them reliably.

Making triggers too generic

Triggers like email, code, or help are too broad. Prefer specific phrases such as draft client email, AI coding prompt, personal voice compiler, or LinkedIn founder post.

---

Positioning

Wispr-style tools personalize transcription. Vocoding personalizes the full workspace: context, output behavior, expert reasoning, reusable skills, and voice intelligence.

Vocoding is not just voice-to-text. It is voice-to-workflow.