Marketing ToolMarketing

Bistec Studio

Write a short brief. A Claude agent writes the captions and designs an on-brand post, and every number is checked against your sources.

About Bistec Studio

The idea

Making a social media post that looks and sounds like the brand usually needs the one or two people who know the brand guidelines and how each platform's publishing works. Everything waits for them. Bistec Studio puts that knowledge into the tool. A brand kit holds the colours, fonts, logos and a versioned brand voice. Anyone on the team fills in a short brief: the campaign, the size, what the post is about and any images. A Claude agent writes the headline and separate Instagram and LinkedIn captions, and designs the post, either by filling a stored brand template or by designing a freeform layout. The design is rendered to a finished image. A facts check then compares every price, percentage, date and number in the headline and captions with the brief, the campaign briefing and documents, and the brand voice, and flags any it cannot find. Editors refine the post in plain words, edit the text inline and keep every version. Admins publish now or schedule, and every publish attempt is recorded. It is BISTEC's own tool, self-hosted with Docker Compose.

Product tour

How Bistec Studio works

  1. Set up the brand once

    A team admin builds a brand kit: colours, fonts, logos, brand templates and a brand voice. The voice is versioned, so an admin can roll it back. Projects and campaigns pick a kit; otherwise the team's default is used.

  2. Write a short brief

    Anyone on the team picks a campaign, a size (1:1, 4:5 or 9:16), a post style and a design path, then says what the post is about and adds any images. The campaign fills in the brand kit and tone. An unfinished brief is saved, so it can be resumed later.

  3. Look over the finished post

    Claude designs the post, and it is rendered to a finished image at the chosen size. The preview shows the size, the post style and which model made it. This sample uses mock data.

    Look over the finished post
  4. Check the headline and each caption

    The headline is the text on the image. Instagram and LinkedIn each get their own caption, with its own character counter, its own Regenerate and a Copy button. If the facts check finds a number it can't trace to the brief or campaign sources, the draft lists it.

    Check the headline and each caption
  5. Refine, and keep every version

    Ask for a change in plain words, like "Make the background darker", edit the text inline, or regenerate the design. Each change becomes a new version, and switching between versions is instant because each keeps its own export.

    Refine, and keep every version
  6. Publish now or schedule

    An admin publishes to Instagram and LinkedIn, each with its own caption, or schedules the post for later. A scheduled post can be cancelled before it goes out. Every attempt is recorded with its time, channel and outcome, and a failure says why.

    Publish now or schedule
The film · 0:38

See Bistec Studio in motion

Sample data · synthetic demo
The problem

What Bistec Studio catches

Case 01

A deadline nobody gave

A programme post says applications close on a set date. The brief never mentioned one; the model filled the gap.

Red flag: The date in the caption appears nowhere in the brief, the campaign briefing or its documents.

Case 02

A salary out of thin air

A hiring post names the role and reads well. It also states a salary.

Red flag: The amount is in none of the sources, and the Hiring style's rules forbid invented pay.

Case 03

A discount that was never offered

A promotion post promises a percentage off until the end of the month.

Red flag: Neither the percentage nor the end date is in the brief or the campaign documents.

Case 04

An off-brand colour

Someone asks the refine chat for a bright red button.

Red flag: Red is not in the brand kit's palette, so the request is flagged as a brand conflict instead of applied.

Stage by stage

How work moves through Bistec Studio

  1. Brand kit

    An admin sets the colours, fonts, logos, templates and brand voice once. A campaign's kit comes first, then its project's, then the team default.

  2. Brief

    Campaign, size, post style, design path, what the post is about and any images. Guided defaults mean no brand or platform knowledge is needed.

  3. Generate

    Claude writes the headline and captions and designs the post from a template or freeform. Puppeteer renders it to a PNG at the chosen size.

  4. Facts check

    Code, not a model, pulls out every price, percentage, date, time, phone number and number, and looks for each one in the sources. Anything not found is listed on the draft.

  5. Refine

    Plain-words requests, inline edits and regeneration, each saved as a new version. A failed render never replaces the current export.

  6. Publish

    Now or scheduled, to Instagram and LinkedIn, each with its own caption. Every attempt is recorded.

Ground rules

Promises it always keeps

  • The brand comes from the kit

    Colours, fonts and logos come from the brand kit and its templates, not from each person's choices.

  • Every value needs a source

    A price, percentage, date or number must appear in the brief, the campaign briefing or documents, or the brand voice. If it doesn't, it is flagged, and a scheduled post set to publish by itself is held.

  • Every version is kept

    Refines, inline edits and regenerations each add a version. You can switch back at any time.

  • Secrets stay on the server

    Claude, image and social credentials are stored encrypted on the server and never reach the browser.

  • Every publish is recorded

    Each attempt keeps its time, channel, outcome and who did it. A failure is recorded with its reason, never dropped.

  • Roles decide who publishes

    Editors create and refine posts. Admins also publish, schedule and look after the brand kits.

Under the hood

What's inside Bistec Studio

Brand kits

Colours, fonts, logos, templates, reference files and the versioned brand voice. Files marked for the AI are given to Claude as brand context.

Brief wizard

Five steps with guided defaults. An unfinished brief is saved and can be resumed from the dashboard.

Design agent

Claude fills a brand template or designs a freeform layout. Background images come from the team's image provider.

Renderer

Puppeteer turns each design into a PNG at 1:1, 4:5 or 9:16, stored in MinIO.

Publisher and scheduler

Publishes to an Instagram Business account and a LinkedIn company page. A separate worker publishes scheduled posts at their time.

Agent access

An MCP server and an HTTP endpoint (ACP) let AI agents create brand kits, generate posts and publish, using a team API key.

By design

What it deliberately won't do

  • It won't: Pixel or canvas editing

    There is no drawing canvas. Changes are made through refine requests, inline text edits or a new design.

  • It won't: Hand designs to an outside editor

    Designs stay in the tool; it does not open them in an external design program.

  • It won't: Make videos

    It makes still image posts and slide decks, not videos.

  • It won't: Publish beyond Instagram and LinkedIn

    Facebook, X, TikTok and YouTube are not supported.

  • It won't: Let an untraced number pass unseen

    A value it cannot find in the sources is listed on the draft, and a scheduled post set to publish by itself is held.

  • It won't: Run as a shared cloud service

    It is self-hosted with Docker Compose, on servers you run.

Technical details

Next.js 14 with TypeScript and React 18, PostgreSQL with Prisma, MinIO object storage, self-hosted sign-in (better-auth), Puppeteer (headless Chromium) to render each design to a PNG, and the Anthropic Claude API or the Claude CLI. Background images come from an image provider the team registers, such as OpenAI. It runs with Docker Compose: the web app, a scheduler worker, PostgreSQL and MinIO. Each person can connect their own Claude token; scheduled work and agent calls use the team's token. Claude, image and social credentials are stored on the server, encrypted with AES-256-GCM, and never reach the browser. Team admins choose which Claude models each step may use (Haiku 4.5, Sonnet 5.5 or Opus 5.5). An MCP server (stdio) and an ACP HTTP endpoint let AI agents generate posts too; both use team API keys that are stored only as a hash and can be revoked at once.

Target audience

Marketing teams, content creators and social media managers who need on-brand posts without a designer or brand expert in the loop. Team admins look after the brand kits, the connected accounts and publishing; super admins manage teams and users and read the usage insights.

Questions

Do I need to know the brand guidelines?

No. The brand kit carries the colours, fonts, logos and voice, and the brief guides you with defaults. That is the point of the tool.

Which channels and sizes does it support?

Instagram and LinkedIn, in 1:1, 4:5 and 9:16. Each channel gets its own caption.

What does the facts check look at?

Prices, percentages, dates, times, phone numbers and other numbers in the headline and captions. Each is looked for in the brief, the campaign briefing and documents, and the brand voice. Anything not found is flagged on the draft, and a scheduled post set to publish by itself is held.

Can I go back to an earlier version?

Yes. Every refine, inline edit and regeneration is saved as a version, and switching between them is instant.

Who can publish?

Editors create and refine posts. Admins also publish, schedule and manage the brand kits. Super admins manage teams and users.

Where does our data go?

It runs on your own servers with Docker Compose, and credentials are stored encrypted. Briefs are sent to the AI providers the team connects, such as Anthropic for Claude and OpenAI for images, which are outside Sri Lanka.

Can AI agents use it?

Yes. An MCP server and an HTTP endpoint (ACP) let agents create brand kits, generate posts and publish, using a team API key that a team admin can revoke at once.

Book a demo

See any of the nine running on your own use case.

A 30-minute walkthrough with the team that built it.

Email us
hello@bistecglobal.com

Offices in Sydney and Colombo.

bistecglobal.com
© 2026 BISTEC Global · Intelligent design. Engineered for adaptability.Demos on this page use sample data.