Codex: следующий шаг после ChatGPT (Полный гайд для НЕпрограммистов)

Summary of technological concepts & product features (Codex, “next step after ChatGPT”)

What Codex is (vs. GPT Chat)

Codex is an OpenAI tool designed to run locally on your computer (on-device) rather than purely as a cloud chat.

Key traits:

• Internet access and the ability to access connected platforms (if you authorize them).
• Uses powerful OpenAI models under the hood.
• Executes tasks described in ordinary language, for example:
  • Sorting files in the Downloads folder.
  • Building a single customer database from multiple disparate files.
  • Creating a visual dashboard from table data without writing formulas/programming.

Architecture (how it works internally)

Codex is described as a 3-part system:

1. Model (“brain”)

   • Understands the task, plans steps, and decides what to do next.

2. Harness / tool layer (“agent tools”)

   • Performs actions such as editing files, opening files, and writing code.

3. Interface

   • Shows what the system is doing via transparent logs of steps, outputs, and file usage (i.e., not “blind automation”).

Sandboxed execution by default

• The agent operates only inside folders you grant access to.
• It shouldn’t touch the rest of the computer unless you explicitly allow it.

Getting started (installation + setup guide)

• Download from opai.com

  • macOS version requires an M-series processor (checked via “About this Mac”).
  • Windows/macOS are supported.

• Login

  • Use a browser login (e.g., Google account) that matches the account used for GPT Chat (redirect linkage).

Project-based workflow

• Create a local folder on your computer.
• Upload/put all relevant files into that folder.
• In Codex, select that folder for the conversation/project.
• Recommended: use separate folders for separate projects.

Interface features & controls

• Settings

  • Includes language selection (noted: Russian).

• Left panel

  • Projects (“threads” grouping multiple chats under one project).

• Model & thinking depth

  • Low: simple/quick answers
  • Medium: default balance of speed vs. depth
  • High / Extra high: complex tasks

• Resolution / access level

  • Default (sandboxed): runs only inside granted folder(s).
  • Full access: may access the entire computer (higher risk; avoid exposing secret keys).

• Upper-right tools

  • Terminal: direct command-line control (likely for programmers).
  • View file changes: shows exactly what the agent modified.

“Skills” (core concept for reusability)

What a Skill is

• A Skill is a reusable action pipeline/instructions—like a digital recipe.
• A skill can be triggered by a phrase to repeat a multi-step workflow.

Implied examples:

• Generating consistent reports
• Creating dashboards
• Tasks like transcription / working with Figm and “Snowtion” (mentioned as built-in examples)

Skill Creator

To turn a successful chat workflow into a reusable skill:

• Create a task (e.g., dashboard from a table)
• Use a command like “/” to access skill tools
• Select Skill Creator
• Codex analyzes the steps taken and packages them into a new skill

Skill editing

• Edit in chat (e.g., “change this, add action”).
• Or edit a text file (Marku format) representing the skill steps.

“Automation” (scheduled skills)

What Automation is

• Automation = scheduled execution of skills (pipeline scheduling).
• Example: run every Monday at 9:00 AM.

Automation workflow described:

• Create automation in the automation tab
• Set the schedule
• Define what to do (e.g., “check data folder; if new files appear, make summary”)

Local-only constraint

• Automation runs only while your computer is on and the Codex app is running.
• It’s not a cloud background job.

Skill Installer + marketplace idea (extensibility)

Skill Installer

• Lets you install skills created by other people.

Setup described

• Copy a skill link (e.g., from GitHub/community catalog)
• In Codex, run skill installer with that link
• Skill downloads/installs and appears in your skill list

Notes

• Some gaps/nuances may require adaptation by the user.

Some nuances/gaps may require adaptation by the user (the developer may leave incomplete steps).

Tutorial / review-style use cases demonstrated

1) Downloads folder cleanup

• Select a Downloads folder.
• Ask in plain language to:
  • Analyze files
  • Sort into categories (documents/images/tables/archives/others)
  • Produce a report of what was moved
• Subtitles claim it completed in “10 minutes” versus much longer manually.

2) Customer database consolidation

• Merge customer/lead data spread across many formats (tables, Excel, CSV, inconsistent naming).
• Standardize fields (e.g., first/last name formatting).
• Find duplicates and fill missing values.
• Output as an Excel-openable table.

3) Analytics dashboard generation

• Input: numeric tables (sales stats, expenses by month, etc.).
• Output:
  • Interactive dashboard
  • Charts for key indicators and period comparisons
  • Anomaly highlighting
  • Export to HTML for browser viewing
• Customization: change color scheme/design.
• Broader suggestion: landing pages / working with web interfaces.

“What to do next” guide (how to get good results)

1. Identify tasks that repeat or waste time.
2. Describe the desired result, not the implementation.
3. Provide context/files and the desired output format.
4. Review, check and clarify until satisfied.
5. Convert the successful workflow into a skill and reuse it.

Main speaker / source

• The subtitles appear to be spoken by a single presenter/author, referred to as “M.” at the end.
• Source video: “Codex: следующий шаг после ChatGPT (Полный гайд для НЕпрограммистов)” (YouTube).