Redacta

Overview

Redacta is a library and CLI tool for generating blog posts, summaries, and enriched documents from video transcripts. It supports multilingual output, image search, and concept illustrations (including ASCII diagrams).

Redacta uses a pluggable LLM transport layer — choose between a fetch-based adapter (browser-safe; works in Chrome extensions and any environment with fetch) or a Copilot SDK adapter (Node.js; uses your GitHub Copilot subscription).

Features

• Generate blog posts and summaries from transcript files (.srt)
• Multilingual support (specify output language)
• Add illustrations: image search and ASCII diagrams to explain concepts
• Outputs in Markdown format with embedded images and diagrams
• Batch processing: Process all .srt files in a directory
• Interactive progress display (TUI)
• Browser-safe library — use createFetchClient in Chrome extensions or web apps
• Pluggable adapters — swap LLM providers without changing application code

Requirements

• LLM access (choose one):
  • GitHub Copilot (authenticated via gh CLI) — used by the CLI by default
  • Any OpenAI-compatible API, Gemini, Anthropic, or HuggingFace — pass an API key via --api-key / --provider-url
• Node.js v18+ (for the CLI; library works in browsers too)
• Google Custom Search API key and Search Engine ID: Required for image search features (--with-illustration).

Installation

npm install -g redacta       # CLI (global)
npm install redacta          # library

GitHub Copilot Setup (CLI default)

The CLI uses @github/copilot-sdk by default. To authenticate:

1. Install GitHub CLI (gh) and Copilot extension: Follow the official guide: Installing GitHub Copilot in the CLI

2. Authenticate: Run gh auth login and ensure you grant access to GitHub Copilot.

Alternatively, skip Copilot entirely and bring your own key with --api-key and --provider-url.

Library Usage

Redacta exports its high-level functions and adapter factories so you can use it programmatically.

Node.js / CLI scripts

import { createCopilotClient, formatTranscript, generateBlogPost } from 'redacta/index.node.js';

const adapter = await createCopilotClient(); // uses gh auth
// or with a custom provider:
// const adapter = await createCopilotClient({ provider: { baseUrl, apiKey } });

const formatted = await formatTranscript(transcript, 'english', 'gpt-4o', adapter);
const blog = await generateBlogPost(formatted, 'english', 'gpt-4o', adapter);

Browser / Chrome extension

import { createFetchClient, formatTranscript, generateBlogPost } from 'redacta';

const adapter = createFetchClient({ provider: 'openai', apiKey: 'sk-...' });
// or Gemini:
// const adapter = createFetchClient({ provider: 'gemini', apiKey: '...' });
// or Anthropic:
// const adapter = createFetchClient({ provider: 'anthropic', apiKey: '...' });

const formatted = await formatTranscript(transcript, 'english', 'gpt-4o', adapter);
const blog = await generateBlogPost(formatted, 'english', 'gpt-4o', adapter);

Supported fetch providers

| Provider | provider value | Custom baseUrl | | :--- | :--- | :--- | | OpenAI | 'openai' | Optional — point to any OpenAI-compatible endpoint (Azure, Ollama, etc.) | | Google Gemini | 'gemini' | — | | Anthropic Claude | 'anthropic' | — | | HuggingFace | 'huggingface' | — |

To use a local Ollama instance or any OpenAI-compatible server, pass baseUrl:

const adapter = createFetchClient({
  provider: 'openai',
  apiKey: 'ollama',           // Ollama ignores the key
  baseUrl: 'http://localhost:11434/v1',
});

Available exports

redacta (browser-safe):

• createFetchClient(config) — creates a fetch-based adapter
• formatTranscript, enrichMarkdown, generateBlogPost, generateSummary
• Types: LLMAdapter, FetchClientConfig, FetchProvider

redacta/index.node.js (Node.js):

• Everything above, plus createCopilotClient(config?) and CopilotProviderConfig

CLI Usage

redacta <transcript_file> [options]
# OR process a directory
redacta --directory <path_to_directory> [options]

Options

| Option | Description | | :--- | :--- | | --blog | Generate blog post (_blog.md) | | --summary | Generate summary (_summary.md) | | --language <lang> | Output language | | --with-illustration | Add essential illustrations | | --with-illustration-all | Add all illustrations | | --directory <dir> | Process all files in directory | | --model <model_id> | Specify model to use | | --search-key <key> | Custom search key for illustration (overrides env) | | --project-id <id> | Custom project id for illustration (overrides env) | | --list-models | List available models | | --api-key <key> | API key for custom LLM provider (BYOK) | | --provider-url <url> | Base URL of custom OpenAI-compatible LLM provider (BYOM/BYOK) | | --provider-type <type> | Provider type: openai, azure, or anthropic (default: openai) | | --provider-wire-api <format> | API wire format for openai/azure providers: completions or responses (default: completions) | | --help | Show help |

Notes

• For illustration (--with-illustration, --with-illustration-all), --search-key and --project-id are required (or set via environment variables CUSTOM_SEARCH_KEY and CUSTOM_SEARCH_PROJECT).
• Use --model to specify the LLM model.
• Use --provider-url and --api-key to bring your own model/key (BYOM/BYOK). Optionally set --provider-type and --provider-wire-api.
• Run with --list-models to see available models.

Examples

1. Format transcript and output in Italian

redacta my_transcript.srt --language italian

2. Format transcript and add essential illustrations

redacta my_transcript.srt --with-illustration

3. Full pipeline: All illustrations, blog post, and summary

redacta my_transcript.srt --with-illustration-all --blog --summary

4. Batch process a directory to Spanish

redacta --directory ./transcripts --with-illustration-all --blog --summary --language spanish

5. Use a custom LLM provider (BYOK/BYOM)

redacta my_transcript.srt --api-key <your_api_key> --provider-url <your_provider_url> --provider-type openai --provider-wire-api completions

Google Custom Search Configuration

To enable image search (--with-illustration or --with-illustration-all), you need Google Custom Search credentials.

1. API Key: Get from Google Cloud Console.
2. Search Engine ID (CX): Get from Google Custom Search Engine.

You can provide these via Environment Variables (recommended) or CLI flags.

Environment Variables:

• CUSTOM_SEARCH_KEY: Your Google API key
• CUSTOM_SEARCH_PROJECT: Your Custom Search Engine ID (CX)

CLI Flags:

• --search-key
• --project-id

Output

Outputs are written as Markdown files in the same directory as the source transcript:

• *_formatted.md: The cleaned and formatted transcript (with illustrations if requested).
• *_blog.md: Generated blog post (if --blog is used).
• *_summary.md: Generated summary (if --summary is used).

License

MIT