Apply AI-based classification to a list of stories based on user's prompts. Uses Redis caching so each story is only classified once

Classify a list of stories using Claude's tool use. Args: prompt_classifier: User-defined prompt (MClassifierPrompt) for classificat

Classify stories using Claude Vision (VLM) — analyzes both text AND images. This is the VLM version of classify_stories_with_ai(). The key differ

* Clear the authentication token

Read the ?sort= comment ordering off the feed URL, defaulting to newest.

Custom sorting for folders. Handles the special case to keep the root folder on top, and also the expectation that despite locale , folders starting w

Create the Elasticsearch index mapping for archives.

Create a imageproxy signed URL for the specified image URL Args: base_url: Base URL of the imageproxy installation hmac_key: HMAC

Detach from the terminal and continue as a daemon.

Delete a folder and unsubscribe from all feeds inside.

Delete a tag from all user's starred stories (stories remain saved, just without the tag). Returns the updated starred counts.

Remove the stored token file.

Returns a copy of the dictionary with all the keys put in a 'do_' namespace

Ensure the Prometheus multiproc directory exists

Used for setting the memcached cache expiry

Return the base36 post id for an individual-post (comments) feed, else None. Reddit serves a single submission's comment thread at URLs like

Reuse the feed's stored tagline so we avoid an extra /about API call.

Return an Atom feed string for this Reddit feed, or None if unfetchable. When the shared minute budget is spent, returns None and sets self.r

Return all classifier notifications for a user, keyed for frontend consumption.

Format a datetime as a human-readable relative string (e.g., '2h ago', 'in 5h 30m').