What the scraper collects
GramClaude's parser collects channels, chats and people into your own bases — through your own accounts, using the requests available to any regular user. There are five modes:
| Mode | Input | Output |
|---|---|---|
| Channel and chat search | keywords — up to 100, plus endings for them | channels and chats from Telegram's global search |
| Similar channels | up to 500 source channels, depth 1–2 | channels Telegram recommends as similar |
| Chat members | up to 200 chats | members, if the chat shows its list |
| Message authors | chats and a message count — up to 50,000 per chat | people who posted, with their message counts |
| Channel commenters | channels, numbers of posts and comments | people who comment on posts |
Keywords for channel search can be suggested by AI from the topic you describe.
What's stored about a person
- ID and username, first and last name;
- interface language, Premium, whether there's a photo and an active story;
- whether they're an admin in the source chat;
- when they were last online — unless they hide it;
- how many messages they wrote in the chat and which keyword matched.
The parser doesn't collect phone numbers, profile bios or location — those fields don't exist in the base. Message text is stored only if you turn that option on.
Filters
By default, bots and deleted accounts are removed. Everything else is opt-in:
- only with a username, only with a photo, only Premium, only admins;
- only with an active story — handy for mass story viewing;
- active in the last N days — a hidden "last seen recently" status doesn't count as inactive;
- excluding accounts Telegram marks as scam or fake.
For channels and chats — member count from and to, whether comments are open, channels only, chats only, or both.
Never collecting the same thing twice
One target written different ways — @username, a t.me link or a post link — counts as one. A person enters the base once; repeats just increase their message count.
The "skip already collected" option checks new people against all your past bases — it's on by default. Results are saved in batches: if a run stops, "Retry" picks up where it left off.
What to do with the base
Export a base to TXT, CSV, JSON or XLSX and choose which fields you need. Inside GramClaude it's immediately available to other modules: people for mass story viewing and direct messages, channels and chats for chat broadcasts, reactions and comments.
Each base has a retention period — 90 days by default, or 30, 180, 365 days or no limit. Expired bases are deleted automatically.
Accounts, limits and FLOOD_WAIT
Scraping runs across your account pool: several accounts work at once and tasks are split between them. Speed is cautious, normal or fast; there's no "no pauses" mode.
Requests count toward the account's read limit — shared by every module that reads, mass story viewing included. If Telegram answers with FLOOD_WAIT, the account rests and another one takes the target; on a long wait the account drops out of the run. What FLOOD_WAIT is and why you can't argue with it — see the separate guide.
What the parser doesn't do
- It doesn't unlock hidden lists. If a chat hides its members, Telegram returns only admins — the parser notices and suggests collecting message authors or commenters instead.
- It doesn't collect channel subscribers. Telegram shows the subscriber list only to channel admins. Commenters are the alternative.
- It doesn't look up phone numbers or tell you who a number belongs to.
- It doesn't add people to chats. GramClaude has no member-adding module.
Telegram's rules and your data
The parser only works with what your account can see as a regular user. Still, Telegram's terms restrict data collection, especially at scale and for AI training, and mass messages to people who didn't expect them lead to spam blocks — see the guide on spam blocks. How you use the collected base, and the responsibility for it, stay with you.
