Comments URL: https://news.ycombinator.com/item?id=50255370

Points: 1

# Comments: 0

Enzyme is 100x smaller and precomputes questions that help agents to write better queries (which theyre getting great at these days), instead of relying on a finetuned model. So it is overall faster

Enzyme produces local artifacts, so technically it could easily run on each team member's machine (it's cheap enough), provided there's another way to sync the knowledge base. Markdown + team kb's is an interesting area and I am also curious to see what standards will take shape -- probably an adjacent problem

I built Enzyme to help agents and people compound knowledge, together. This week I decided to open-source several years of my work (it began as an Obsidian plugin, then became the first MCP for Obsidian).

Enzyme is like an LSP for ideas; it orients agents by emulating your working memory from a local knowledge base (sqlite or .md). I decided that a language was the best way to represent several design elements of agent memory at once: append-only log locations, temporal distribution config, profiles for "dreaming", and more. Your agent can setup Enzyme, compile the Enzyme program, and maintain a local-first, self-updating lens over its wiki. It's very fast.

I use it daily to maintain my 8000+ note Zettelkasten that also acts as a personal CRM (meeting transcriptions), Readwise collection, journal, and home for agent rollout summaries.

Core to its design are questions that are generated from recent material and projected over the corpus. Instead of "dreaming" .md files over fragments, it generates questions to connect their ideas using embedding similarity (static embeddings developed by Flower Computer). What these questions notice is configurable.

The binary, including the embedding weights baked in, sits under 30 mb.

Comments URL: https://news.ycombinator.com/item?id=49744733

Points: 1

# Comments: 0

aside is a ~5MB Rust binary that records meetings and feeds them into an AI-native transcription-to-vault workflow:

1. Record — captures mic + system audio simultaneously (stereo WAV, left=you, right=them) with a timestamped TUI editor for real-time notes 2. Transcribe — local transcription via whisper.cpp, with a 7-stage cleanup pipeline that strips hallucinations, deduplicates backchannels, and merges fragments 3. Align — interleaves your memo lines with the transcript on a shared timeline, so you can see what was said around each thing you noted 4. Distill — a Claude Code skill searches your Obsidian vault for related notes, then writes a structured note back into it with [[wikilinks]] to your existing thinking

The distillation step is the part I'm most interested in feedback on. The entire pipeline is a 240-line markdown file (a Claude Code "skill") that the LLM follows as instructions. It treats your memo lines as attention signals — what you wrote down mid-call gets priority over what you didn't. Lines you edited mid-meeting get weighted even higher. It searches your vault semantically and by structure (tags, people, wikilinks), then weaves connections into the output note.

The output doesn't read like a meeting summary with action items. It reads like what you would have written up yourself if you had the time — highlighting what mattered to you, connected to what you've already been thinking about.

The skill is plain text. You can read it in the repo, fork it, change how it prioritizes topics or formats notes. There's no black box between the transcript and the vault note — just markdown instructions an LLM follows.

This matters because the workflow is vault-native end to end: sessions live inside the vault, distilled notes land where your thinking already lives, and the AI step uses your vault's own language and structure as context. The meeting note isn't a standalone summary — it's a node in your existing knowledge graph.

Technical details: 3.1k lines total. Rust binary handles recording + TUI (cpal for audio, ratatui for the editor, lock-free ring buffers for real-time capture). Python script handles transcription cleanup. Claude Code skill handles distillation. Everything runs locally except the optional LLM call.

Dual audio capture solves hybrid meetings — conference room mic + remote participants on speaker both get transcribed on separate channels.

No accounts, no telemetry, Apache 2.0. Install via `brew install jshph/aside/aside`

Comments URL: https://news.ycombinator.com/item?id=47233839

Points: 1

# Comments: 0

having worked on the AI Playlist feature at Spotify (which is agentic) -- one thing you can try to prompt is "give me songs that sound from my .

An extension of this is "give me a sequence of N tracks from each of that all evoke ". Could be a fun prompt to play with

I'd hazard a guess that from the writer's perspective, novelty scales with volume of thought / connections, which is (at present) a fragmented process and not that well-assisted by AI. OTOH, can "writing quality" be better approximated by LLMs?

The goal of plumbing is to fix / repair; certainly it's possible to enjoy fixing and repairing. But is the joy of writing in "repairing ideas"? How is that a separate concept from creating new ones?