Nalgorithm — Nostr Relevance Feed

Settings

Identity

Connected

Reads your public key only — Nalgorithm never signs or posts.

npub or hex pubkey Remote signer relays (one per line)

Used only for the Amber / bunker handshake. Must relay ephemeral kind 24133.

Relays

WebSocket relay URLs (one per line)

LLM API

Provider API Base URL API Key

Model (scoring)

Does the bulk of the work. Keep it cheap and fast.

Model (digest)

Runs once over your top posts, so a stronger model costs little. Blank reuses the scoring model.

Model (learning from likes)

Needs a large context window. Blank reuses the scoring model.

Digest

Posts to include Phrase it for speech

Drops markdown and spells out version numbers and abbreviations, so it reads well aloud.

TTS model (for downloadable audio) TTS voice Read aloud with

Browser speech is free, instant, and never leaves the device — the voice is whatever your OS ships. A TTS provider sounds far better and costs a little. With no TTS model set, browser speech is used either way.

TTS API Base URL TTS API Key

Routstr does not sell speech — its catalog is text models only — so if you score with Routstr, point these at a provider that does (Venice: https://api.venice.ai/api/v1 with tts-kokoro, about 2.7 cents a digest). Leave blank to reuse the LLM provider.

Prompts

User prompt (describe your interests) Learned prompt (auto-generated from likes)

Feed

Time window (hours) Batch size (posts per LLM call) Parallel scoring batches

How many scoring calls run at once. 1 is sequential; 5 is roughly five times faster on a cold cache. Keep it within what your provider's rate limits allow.

Open posts in

Use {e} where the event id goes, or just give the prefix — both work.

Load the feed automatically on open

Connect your Nostr identity and add an LLM API key in Settings, then click Refresh to load your feed.

Everything runs in your browser. Your key and settings stay in this device's local storage.