Keep personal rules available
Use Philosophy for durable instructions, taste, review standards, decisions, and boundaries that survive one chat.
Checking model service...
Personal AI assistant plan
Clauxel Console is the working surface above this guide. Its personal AI assistant is built from five visible modules: Philosophy, Knowledge, MCP, Skills, and LLM. Together they preserve context, retrieve private sources, approve tools, repeat workflows, and route each task to the right model.
Direct answer
It should turn a request into a controlled personal workflow. The AI agent console uses saved preferences, retrieves private knowledge only when relevant, asks before tool actions, follows Skills for known procedures, and keeps LLM routing visible.
Use Philosophy for durable instructions, taste, review standards, decisions, and boundaries that survive one chat.
Use Knowledge for notes, documents, and references, then separate grounded answers from assumptions.
Use MCP connections for files, browser context, APIs, and services, with approval for sensitive actions.
Use Skills so research, writing, testing, publishing, review, and reporting follow a consistent path.
Operating loop
The page above is the command surface. The loop below shows what should happen before it returns a result you can trust.
Five modules
A personal AI assistant becomes reliable when memory, evidence, tools, procedures, and model choice are visible inside Clauxel Console.
Philosophy is durable memory for principles, writing taste, review rules, preferences, and decisions the console should remember.
Read the Philosophy guide Module 02Knowledge is the private source layer. Keep notes, documents, saved context, and imported material available for retrieval with visible source boundaries.
See the Knowledge layer Module 03MCP is the connection layer for tools and context. Use it when a task needs controlled access to files, browser state, APIs, or external services.
Explore MCP resources Module 04Skills are the repeatable workflow layer. A Skill tells the console how to gather context, apply a rubric, use tools, verify, and report.
Browse Skills Module 05LLM is the model layer. The composer exposes built-in models, reasoning, speed, and saved custom models for the same personal workflow.
Choose an LLM routeCapability map
Use it for recurring research, documents, code checks, websites, scheduled operations, and saved-context decisions.
| Personal need | Module | What good behavior looks like |
|---|---|---|
| Remember how I work | Philosophy | The AI agent console uses durable preferences and lets the owner edit or remove stale rules. |
| Answer from my material | Knowledge | The AI agent console retrieves relevant sources and marks uncertainty when the vault is incomplete. |
| Use a tool without losing control | MCP | The AI agent console explains the action, asks when needed, and leaves an inspectable tool log. |
| Run the same workflow again | Skills | The AI agent console follows a named procedure, verifies output, and reports changes. |
| Switch models for the job | LLM | The AI agent console keeps built-in and custom models visible in the composer. |
| Keep the task personal | Private Assistant | The AI agent console treats projects as organization aids, not as the owner. |
Customer journey
The journey should be obvious: understand Clauxel Console, add context, run a task, approve actions, review output, and improve the next run.
The first screen shows the composer, modes, model picker, side modules, account state, upgrade status, and feedback. The guide shows what it can remember, retrieve, connect, run, and route.
Result: personal, not generic.
The user adds Philosophy for durable rules and Knowledge for private sources. The AI agent console treats them separately: one shapes behavior, the other grounds answers.
Result: preferences and evidence are separate.
The user sends a prompt, attaches files, or chooses Deep Research, Docs, Code, Slides, Websites, Sheets, or Swarm. The AI agent console keeps the mode visible.
Result: less prompt repetition.
If the task needs an MCP tool, account connection, file action, API call, schedule, or external effect, the AI agent console explains it first.
Result: sensitive work stays controlled.
The user selects a built-in model, reasoning and speed, or a saved custom model. The selected model stays visible in the composer.
Result: routing is easy to confirm.
The final answer gives the result, cites Knowledge when used, shows tool traces, and makes it easy to save a better Philosophy or Skill.
Result: the next run is easier.
Boundaries
The console should help without silently crossing important lines. Clear boundaries are part of the product.
| Action type | Expected behavior | User-visible check |
|---|---|---|
| Read saved context | Use Philosophy and Knowledge when relevant to the task. | Show when private knowledge influenced an answer. |
| Write a new memory | Suggest the Philosophy or Knowledge update before saving it. | Let the owner review, edit, or reject the update. |
| Call an MCP tool | Explain the tool, input, and effect before sensitive actions. | Display permission prompts and tool logs. |
| Use a custom model | Route only through models saved in LLM or selected in the composer. | Keep the active model name visible. |
| Spend money or change accounts | Require owner approval before payments, credentials, deletion, or account-sensitive changes. | Never hide these actions inside a normal answer. |
A dependable console should not pretend. If a source, tool, credential, model, or approval is missing, it names the gap and gives the owner a next step.
Task template
A good task still needs a goal, allowed context, tool boundary, output shape, and acceptance check.
Goal: ...
Use my personal AI assistant.
Philosophy: ...
Knowledge: ...
MCP tools: ...
Skills: ...
Model: fast / balanced / careful / custom
Human approval: payments, accounts, credentials, deletion, public posting, external services.
Output: ...
Done when: ...Before pressing send, know what changes, which context matters, which tools are allowed, and what proof is enough. If anything is missing, it asks instead of guessing.
Examples
Use Clauxel when work needs continuity across chats: stored preferences, private sources, tool permissions, a known method, and flexible model choice.
Save assumptions in Philosophy, pull notes from Knowledge, and ask the console for a brief that separates known facts from uncertainty.
Give the console style rules, attach sources, choose careful reasoning, and use a Skill for copy, notes, or review.
Use MCP connections and schedules to check a page, prepare a document, organize a sheet, or follow up with context intact.
Run the same task on a fast model, a careful model, and a saved custom model, then compare output quality.
Open interfaces
Clauxel keeps the console understandable: memory for durable context, tools for action, skills for procedures, and routing for model choice.
FAQ
Use these boundaries before a first personal task.
Yes. Projects can organize a topic, but the console stays centered on one owner, private knowledge, permitted tools, skills, and preferred models.
Philosophy is preferences, review rules, boundaries, and working patterns. Knowledge is source material such as notes, documents, references, and imported context.
Use MCP when the console needs files, browser state, APIs, databases, or services. Sensitive actions stay visible and permissioned.
Skills package procedures, checklists, files, and verification habits so the console can run familiar work consistently.
Yes. The LLM panel saves custom model configuration; configured models appear in the composer model picker.
Money, credentials, deletion, account checks, and external service changes need owner permission.
Ready when the context is ready
Start with one task, then save what should persist: a Philosophy, Knowledge source, MCP permission, Skill, or custom model. The goal is an AI agent console that becomes clearer, safer, and more useful each time you return.