Telegram → Discourse Content Engine

A Telegram → Discourse content engine built as a chat-first claws-and-skills system, with both Dockerized execution and an OpenClaw-compatible control surface.

The repository supports two complementary usage modes in the same codebase:

• Dockerized runtime mode — run the content engine and database in containers
• OpenClaw-compatible mode — let an external OpenClaw installation drive this repo through stable machine-friendly actions

This repository demonstrates a clean separation between:

• Claws — lightweight workflow orchestrators
• Skills — reusable operations that can be used independently of this project

The system is incremental by design: it processes only new Telegram posts, stores extracted knowledge in PostgreSQL with pgvector, schedules future writing tasks, generates AI drafts, and publishes ready drafts.

Fastest Launch

From the workspace root:

chmod +x ./content-engine.sh
cp openclaw-content-engine/.env.docker.example openclaw-content-engine/.env.docker
./content-engine.sh docker setup
./content-engine.sh docker chat

The root content-engine.sh script is the recommended entrypoint. For open-source usage, the simplest path is the Docker-first flow above. If you want to run everything directly on your machine instead, use the local commands further below.

---

Docker Setup

The workspace includes a containerized runtime for isolated execution:

• Dockerfile — builds the multi-repo Node.js workspace
• docker-compose.yml — runs the content engine and PostgreSQL with pgvector
• .env.docker.example — example container environment file

./content-engine.sh docker db
./content-engine.sh docker analyze
./content-engine.sh docker schedule
./content-engine.sh docker write
./content-engine.sh docker publish
./content-engine.sh docker action plan.show
./content-engine.sh docker stop
./content-engine.sh docker reset

What they do:

• docker setup — build the image and start the database
• docker chat — launch the chat interface
• docker db — start only the database
• docker stop — stop the Docker stack
• docker reset — stop the stack and delete the Docker database volume

---

OpenClaw-Compatible Mode

This repo exposes a stable action interface intended for an external OpenClaw controller.

Compatibility assets:

• prompt.md — domain prompt for the content engine agent
• openclaw/actions.json — machine-readable action catalog
• node dist/index.js action ... — stable action execution CLI

Example host usage:

node dist/index.js action plan.show
node dist/index.js action plan.schedule '{"humanComment":"Prioritize security and infra topics","maxItems":3}'
node dist/index.js action article.write '{"topicName":"AI agent economy"}'

Example Docker usage:

docker compose run --rm engine node dist/index.js action plan.show
docker compose run --rm engine node dist/index.js action article.publish

This gives you one repo with two operating styles:

• self-contained Docker execution
• external OpenClaw-driven execution

---

Core Architecture

openclaw-content-engine/
  src/
    claws/
      content-engine-claw.ts
      telegram-analyzer-claw.ts
      publication-scheduler-claw.ts
      article-writer-claw.ts
      article-publisher-claw.ts
    skills/
      telegram-reader-skill.ts
      article-extractor-skill.ts
      semantic-utils-skill.ts
      topic-memory-skill.ts
      editorial-intelligence-skill.ts
      discourse-publisher-skill.ts
    runtime.ts
    index.ts

Claws vs Skills

Claws

Claws are orchestration units. They coordinate multi-step workflows, persist progress, and decide which skills to call next.

Implemented claws:

• ContentEngineClaw
• TelegramAnalyzerClaw
• PublicationSchedulerClaw
• ArticleWriterClaw
• ArticlePublisherClaw

Skills

Skills are reusable operations. They wrap workspace support packages and expose clean project-facing interfaces.

Implemented skills:

• TelegramReaderSkill → wraps the workspace package @openclaw/telegram-channel-reader
• ArticleExtractorSkill → wraps the workspace package @openclaw/article-extractor
• SemanticUtilsSkill → wraps the workspace package @openclaw/semantic-skills
• TopicMemorySkill → wraps the workspace package @openclaw/topic-memory-db
• DiscoursePublisherSkill → wraps the workspace package @openclaw/discourse-api-client
• EditorialIntelligenceSkill → provides OpenAI-backed planning and writing

The @openclaw/* package scope here is this workspace's internal package namespace. It is not meant to imply that these packages come from an official OpenClaw SDK.

---

Chat-First Control Flow

OpenClaw Chat Interface
  ↓
ContentEngineClaw
  ↓
Specialized claw actions
  ├─ TelegramAnalyzerClaw.analyze()
  ├─ PublicationSchedulerClaw.showPlan()
  ├─ PublicationSchedulerClaw.schedule()
  ├─ ArticleWriterClaw.write()
  ├─ ArticleWriterClaw.writeAbout()
  └─ ArticlePublisherClaw.publish()

The user interacts with the content engine through chat. ContentEngineClaw is the primary control mechanism and decides which specialized claw action to call.

For machine-driven integrations, the same action surface is also available through node dist/index.js action <name> '<json-args>'.

Example interactions:

• Analyze new Telegram posts
• Show the current content plan
• Write an article about the AI agent economy
• Publish scheduled articles

---

End-to-End Pipeline

Telegram channel
  ↓
TelegramAnalyzerClaw
  ↓
Posts + extracted articles + embeddings + topic links
  ↓
PublicationSchedulerClaw
  ↓
Content plan items
  ↓
ArticleWriterClaw
  ↓
Persisted draft articles
  ↓
ArticlePublisherClaw
  ↓
Published Discourse topics

1. TelegramAnalyzerClaw

Responsibilities:

• read new channel posts using telegram-channel-reader
• detect URLs
• extract article content using article-extractor
• generate embeddings via semantic utilities
• cluster related content into topics
• store structured knowledge in topic-memory-db

Output:

• stored posts
• stored articles
• stored embeddings
• topic/article links
• structured article knowledge objects

2. PublicationSchedulerClaw

Responsibilities:

• review stored articles and topics
• detect which topics are already covered
• cluster similar material for editorial prioritization
• use the editorial AI skill to rank candidate topics
• create content plan entries with optional human comments

Output:

• scheduled article tasks in content_plan

3. ArticleWriterClaw

Responsibilities:

• take an approved scheduled topic
• retrieve related source articles and summaries
• incorporate human comments
• generate a concise article and tags using the editorial AI skill
• persist drafts in draft_articles
• update plan item status

Output:

• stored article drafts ready for publication review

4. ArticlePublisherClaw

Responsibilities:

• retrieve ready drafts
• format and publish them to Discourse
• persist publication metadata
• mark draft and plan statuses as published

Output:

• published Discourse topics
• updated pipeline state

---

Supporting Repositories Used as Skills

Package	Purpose
telegram-channel-reader	Incremental Telegram post retrieval
article-extractor	URL → clean article extraction
semantic-skills	Embeddings, cosine similarity, clustering
topic-memory-db	PostgreSQL + pgvector knowledge store
discourse-api-client	Discourse publishing

These repositories remain reusable workspace libraries. The OpenClaw-compatible binding lives in this repo’s skills/ layer and CLI action surface.

---

Persistent Data Model

Long-lived editorial memory is stored in PostgreSQL.

Table	Purpose
posts	Telegram messages with URLs and metadata
articles	Extracted article content, titles, summaries
embeddings	vector(1536) embeddings with IVFFlat index
topics	Discovered topic clusters with centroid vectors
topic_articles	Many-to-many links between topics and articles
content_plan	Editorial plan items
draft_articles	AI-generated article drafts persisted before publishing
published_articles	Published Discourse topic/post references
pipeline_state	Key-value store for incremental processing state

Key properties:

• native vector(1536) embeddings
• pgvector cosine search support
• incremental last-processed Telegram state
• durable AI draft persistence

---

Execution Model

Primary interface:

• ./content-engine.sh docker chat

Additional machine-friendly interface:

• node dist/index.js action <action-name> '<json-args>'

Convenience commands:

• ./content-engine.sh docker analyze
• ./content-engine.sh docker schedule
• ./content-engine.sh docker write
• ./content-engine.sh docker publish
• ./content-engine.sh docker run
• ./content-engine.sh analyze
• ./content-engine.sh schedule
• ./content-engine.sh write
• ./content-engine.sh publish
• ./content-engine.sh run

run may still be used, but it is only a convenience wrapper that triggers a sequence of chat-routed claw actions through ContentEngineClaw.

Example convenience sequence:

1. TelegramAnalyzerClaw processes new posts
2. PublicationSchedulerClaw updates the content plan
3. ArticleWriterClaw generates drafts for approved tasks
4. ArticlePublisherClaw publishes ready drafts

CLI commands:

./content-engine.sh docker chat
./content-engine.sh docker analyze
./content-engine.sh docker schedule
./content-engine.sh docker write
./content-engine.sh docker publish
./content-engine.sh docker run
./content-engine.sh analyze
./content-engine.sh schedule
./content-engine.sh write
./content-engine.sh publish
./content-engine.sh run

You can still invoke the orchestrator directly with node dist/index.js ..., but the root launcher is the recommended entrypoint for this multi-repo workspace.

---

Setup

Requirements

• Node.js ≥ 18
• PostgreSQL ≥ 15 with the pgvector extension
• Telegram API credentials from my.telegram.org
• OpenAI API key
• optional Discourse API credentials

Install

Recommended host setup:

./content-engine.sh setup

If you do not want to install dependencies on your host, use the Docker workflow documented above instead.

Database

createdb openclaw
psql openclaw -c "CREATE EXTENSION IF NOT EXISTS vector;"

Environment

See .env.example for the full list:

• TELEGRAM_API_ID, TELEGRAM_API_HASH, TELEGRAM_CHANNEL — Telegram access
• OPENAI_API_KEY — embeddings, scheduling, and article generation
• DATABASE_URL — PostgreSQL connection string
• DISCOURSE_URL, DISCOURSE_API_KEY, DISCOURSE_USERNAME — Discourse publishing

For ./content-engine.sh chat, analyze, schedule, or write, only the Telegram/OpenAI/DB values are required. For ./content-engine.sh publish or run, publish credentials are required.

---

Design Principles

• Claws stay lightweight
• The chat agent is the primary control interface
• Skills remain reusable
• AI is used for planning and writing, not just deterministic routing
• Incremental operation is mandatory
• Persistent semantic memory prevents wasteful reprocessing
• Open-source readability matters

Integration Status

What is implemented today:

• a chat-first control layer via ContentEngineClaw
• specialized claws for analysis, scheduling, writing, and publishing
• reusable workspace packages under the @openclaw/* scope
• Dockerized isolated execution for the app and database
• an OpenClaw-compatible action entrypoint for external control

What this means in practice:

• you can run the system directly through Docker or the local launcher
• you can let an external OpenClaw setup drive this repo through node dist/index.js action ...
• the compatibility surface is explicit and machine-friendly

What is not implemented yet:

• a verified official OpenClaw SDK dependency in this codebase
• official OpenClaw runtime registration or packaging conventions from docs.openclaw.ai

---

Status

This repository now serves as a dual-mode content engine: self-hosted through Docker or the local launcher, and externally controllable through an OpenClaw-compatible action interface.