Automating Telegram content with multi‑agent system
Hi
What we’re building
An AI‑driven Micro SaaS that runs themed Telegram channels end‑to‑end using a user’s “digital twin” (style, tone, interests):
Research: generates focused web queries based on the channel’s profile; fetches fresh sources.
Curation: ranks/dedupes content; filters low‑signal links.
Writing (Reasoning LLM): produces short, engaging posts in your voice (first‑person OK), with optional A/B style overlays.
Quality gate: similarity vs channel history (to avoid repeats) and readability check.
Publishing: schedules and posts automatically (or on demand) with a single bot across multiple channels.
Analytics + feedback: auto‑learns preferred keywords/length; optimizes posting times.
Why this is Micro SaaS‑friendly
Narrow scope: “Run my Telegram channel in X niche” (e.g., crypto, tech, games).
Low overhead: lightweight stack.
Clear value: consistent posting, tone match, measurable engagement.
Repeatable playbook: onboard a new niche → set sources/keywords → go.
Who it’s for
Solo creators, niche brands, and community owners who want consistent, on‑brand posts without hiring a team.
Agencies who need white‑label automation for multiple client channels.
Tech snapshot (for the curious)
Multi‑agent pipeline (Research → Curate → Write → QC → Schedule/Publish → Analytics).
Reasoning LLM, web search/extract, Telegram Bot API, embeddings for similarity.
Deterministic scheduling with manual “generate‑now”.
Pricing ideas
Subscription per channel (with usage tiers) or agency bundles. Optional add‑ons: source whitelists, custom prompts, A/B testing, analytics exports.
Looking for early users
If you run a Telegram channel (or a few) and want to trial a “set‑and‑forget” workflow with a sarcastic/serious/analytical voice that actually fits your brand, DM me your niche and target cadence. We’ll spin up a digital twin + content preferences and get your first posts live this week.

Replies
@derek_brambles
do you mean tech?
python, postgres, redis, openai and antropic
@derek_brambles
Here’s how we handle images (and “alt”):
We don’t have true HTML alt in Telegram. We treat the photo caption as the “alt” and, for accessibility, optionally send a second text message with a fuller description right after the image.
Captions are generated from context with a vision model and lightly normalized. For purely decorative images we use an empty/very short caption.
We cache and reuse Telegram file_id to avoid re-uploads; if quality is critical we send as document (keeps original) instead of photo (compression).