PKM (BASB, Zettelkasten, or "Digital Garden") integration

[TomLucidor]

There are a lot of options out there for PKMs to cache aggregated knowledge (in order of star count)

• LobeChat is purely chatbot-centric https://github.com/lobehub/lobe-chat
• Affine is closer to Notion https://github.com/toeverything/AFFiNE
• LogSeq is a FOSS classic closer to Obsidian https://github.com/logseq/logseq
• Outline is another Notion clone https://github.com/outline/outline
• Siyuan is a new player with LLM integration https://github.com/siyuan-note/siyuan

Some smaller names include:

• Trilium is moving on to TriliumNext https://github.com/TriliumNext
• Foam and Dendron are VSCode-based https://github.com/foambubble/foam https://github.com/dendronhq/dendron
• DocsGPT and Onyx are other chatbot centric tool https://github.com/arc53/DocsGPT https://github.com/onyx-dot-app/onyx
• Reor is very minimalist with LLM integration, cross-linking support and other features are WIP https://github.com/reorproject/reor

Some of the core considerations with TLDW:

• Can it read source text and media files? And cache them into semantic/vector DB?
• Are churning out new text files in the PKM using LLM possible? (Bonus: tracking human vs LLM work)
• Will it handle page cross-links when remixing new ideas together? ([[page]] for LogSeq and Obsidian)
• How can human feedback be integrated in the knowledge-consolidating process?
• What about disambiguation and synonyms?

[rmusser01]

I feel like this is two separate things,

1. Comparisons to other chat frontends/PKM parallel tooling;
2. Clarification of feature goals for tldw;

To 1, Yea, there's a lot of them and I would like to take a deeper look at each to figure out what features I'd like to copy/use/modify for my own.

To 2,

1. Yes
2. Not yet but is planned, the idea is to allow for 'synchronizing' notes between DBs (starting with Obsidian as a PoC)
3. Not currently planned, but likely to be added later. The current DB implementation for notes is specifically just what's needed for current functionality. Future updates would likely implement this.
4. Options for checkpoints when performing analysis as well as feedback (thumbs up/down) ranking on results can help the user create a curated selection of results to help in further finetuning of their model of choice.
5. This is being added soon*(today/tomorrow depending) through the use of chat dictionaries, which will allow the user to perform replacements on existing words/acronyms in the pre-submitted prompt and the returned prompt response. This should also tie nicely into the recent popularity of reasoning models and R1/QwQ 'thinking' processes.

[TomLucidor]

Regarding core point 1, yeah if there are ways to pick the most popular FOSS ones for support (cus LogSeq/Siyuan > Obsidian), + some indie ones like Reor, it would be more useful then just starting from scraps.

• As for no.2 it is more like using LLM agents to help with summarization and preliminary work
• For no.4 it would take some work figuring how to keep track of human preference as the core note-box mutates beyond the original