Telegram bot message persistence

[Teem-git]

Hi so i connected my telegram bot to my running fcc instance but everytime i send a message it uses a new CLI instance how do i make it have memory

[atokim4]

Reply to the bot's message instead of sending a new one. In Telegram that's long-press (or swipe left) on the bot's answer and pick Reply, then type your follow-up.

FCC keeps each chat as a tree of messages. A message that isn't a reply starts a new tree, and you'll see "Launching new Claude CLI instance..." for it. When your message is a reply, it looks up the message you replied to (your earlier prompt or the bot's status/answer message both work), takes the Claude session id saved on that node and runs the CLI with --resume <that session> --fork-session. The status then says "Continuing conversation..." and Claude has the earlier context. You can see this in messaging/turn_intake.py (the is_reply() branch) and cli/managed/claude.py where the resume args are built.

Because it forks, you can also reply to an older message to branch off from that point while keeping the other thread. If you reply while the earlier turn is still running, it gets queued and continues once that finishes. Standalone /clear drops all conversations in the chat; replying /clear to a message only removes that branch.

I checked this against the current main and ran the repo's handler tests for the fork path, with an AI assistant helping me trace the code.

[Teem-git]

Thank you for the prompt response I now better understand the how to use this feature, another question is it possible for it to read images

[atokim4]

Not from Telegram directly right now. In messaging/platforms/telegram.py the bot only registers handlers for text, commands and voice notes (filters.TEXT, filters.COMMAND, filters.VOICE), so a photo you send just gets ignored.

What you can do instead: put the image inside your Allowed Directory (for example screenshots/error.png) and send a text message like "look at screenshots/error.png and tell me what's wrong". Claude Code can open image files with its Read tool. Whether it actually understands the picture depends on the model you route to, since it has to accept image input. A text-only model won't see anything useful.

If you want native photo support, that would need a filters.PHOTO handler that downloads the file into the workspace, similar to how the voice flow saves audio. That could be worth a feature request.

(Same as before, I checked this against current main with an AI assistant helping me read the code.)