Context Prism: reduce AI token waste with local context packs

Share & showcase
,
1

Is this project open source? Yes
Is this project completely free? Yes
Is this project vibe-coded beyond the author’s ability to comprehend how it works? No
Community Directory: Context Prism - Obsidian Plugin

Hi everyone, I built Context Prism, a local-first plugin that creates token-aware context packs from related notes.

The problem I wanted to solve is simple: when using AI assistants with a large vault, it is very easy to paste too much context. That wastes tokens, makes prompts harder to inspect, and often gives the assistant notes that are not actually useful.

Context Prism ranks related notes locally from the active note and creates a compact Markdown pack with:

  • note paths
  • focused snippets
  • ranking reasons
  • rough token estimates
  • estimated avoided context

You can paste that pack into ChatGPT, Claude, Codex, Cursor, or any other assistant before asking for analysis, writing help, planning, or code work.

It does not call external APIs, does not send note content anywhere, supports multilingual vaults, and also includes a manual link review workflow if you want to insert wiki-links.

Install:

I am especially looking for feedback on real AI workflows:

  • Does it surface the notes you would have copied manually?
  • Does it miss important context?
  • Are irrelevant notes still appearing too high?
  • Are the snippets useful enough for the assistant?
  • Do the token estimates help you decide what to paste?
  • What features would make this more useful in your own vault?

If you try it, I would really appreciate examples of where the ranking works well, where it fails, and what kind of vault/workflow you tested it with. Synthetic or redacted examples are ideal.

Thanks. I would like to use community feedback to improve retrieval quality, multilingual behavior, and future context-pack features.

context-prism-demo

2

Small update: I realized my original post was written more like an announcement than a feedback request.

What I actually need now is narrower: I am trying to understand whether Context Prism surfaces the notes people would manually give to an AI assistant.

I created a pinned 3-minute feedback issue in the GitHub repo. You can find it as:

“Early feedback: did Context Prism surface useful notes?”

If anyone tries it, the most useful feedback would be:

  1. What kind of note did you test it on?
  2. Did it surface notes you would have copied manually?
  3. What was the biggest issue: missing notes, irrelevant notes, snippets, token estimates, or workflow?

No private notes needed. Synthetic examples or a short redacted description are enough.

3

This is a problem I used to run into. I open Claude or Copilot and start re-explaining my whole project, and halfway through I remember I already wrote something better in my vault — but pulling it in by hand is more friction than just typing it again from memory.

Right now, I have a dozen of files that provide context for most of my needs.

What does your ranking actually key on? I’m curious whether you rank on semantic text similarity, or whether you also use the note structure — backlinks, outgoing links, the graph — because that is where I’ve seen these tools diverge. Text similarity finds notes that talk about the same thing, but the link graph often finds notes that complement each other.