Skip to content

product-positioning.md

Latest commit

 

History

History
143 lines (98 loc) · 4.67 KB

product-positioning.md

File metadata and controls

143 lines (98 loc) · 4.67 KB
Raw

Product Positioning

Kosmos is the local trust layer for AI working inside Obsidian-style vaults, repos, prompts, and knowledge bases.

It should not compete with Obsidian as a writing app. Obsidian is where people write, link, and navigate their notes. Kosmos should sit beside it and answer the question Obsidian does not answer on its own:

Can I trust what my AI just did to this workspace?

One-Sentence Pitch

Kosmos shows what AI agents read, changed, cited, missed, or broke in your local vault or repo.

Audience

  • Obsidian users letting AI assistants operate on their vaults
  • developers using Claude Code, Codex, Cursor, OpenClaw, or local agents against a repo
  • researchers and builders turning raw notes into generated outputs
  • prompt and agent builders who need replay, provenance, and context health without a hosted dashboard

The Problem

AI tools are increasingly allowed to read, edit, summarize, reorganize, and generate from local context. Once that happens, users lose track of:

  • which files the agent actually read
  • which notes, prompts, and instruction files shaped the answer
  • which outputs are grounded in sources
  • which links, citations, or assumptions broke
  • which generated artifacts are stale
  • which changes came from a human versus an AI session

The pain is not "I need another markdown editor." The pain is "I need evidence before I trust this AI-assisted workspace."

What Kosmos Owns

Kosmos should own agent accountability over local context:

  • context graph: notes, sources, outputs, prompts, instructions, tools, and files
  • live activity: what an AI tool is reading, writing, calling, and producing
  • replay: what happened in a session, in order, with file evidence
  • provenance: what sources support this output
  • context health: broken links, missing sources, orphan notes, stale outputs, risky instructions

What Kosmos Does Not Own

Kosmos should not try to replace:

  • Obsidian as the primary writing and note navigation environment
  • VS Code or other editors as the primary coding environment
  • hosted production observability platforms for large SaaS telemetry
  • full project management, CRM, or document management workflows

Product Language

Prefer:

  • Trust Overview
  • Context Map
  • AI Sessions
  • Context Audit
  • Source Coverage
  • Unsupported Outputs
  • Agent-Touched Files
  • Broken Links
  • Missing Sources

Avoid leading with:

  • generic dashboard
  • universe map as the main promise
  • architecture observability
  • AI operating system
  • markdown editor
  • note-taking app

GitHub Star Wedge

The star-worthy wedge should be fast, local, and easy to demo:

npx kosmos-aos ~/Obsidian/MyVault
npx kosmos-aos ./path/to/your-vault
npx kosmos-aos ./examples/ai-context-vault

Then Kosmos should immediately show:

  • AI readiness score for the vault or repo
  • prioritized review queue for the worst context issues
  • broken wikilinks
  • pages missing sources
  • orphan notes
  • generated outputs without provenance
  • instruction files that shape AI behavior
  • recent AI sessions and touched files when live ingest is connected

The emotional payoff should be:

I can let AI work in my vault without losing the plot.

Product Changes Needed

1. Make Obsidian-style vaults first-class

  • detect .obsidian/
  • parse wikilinks, aliases, tags, frontmatter, and source fields
  • show an explicit vault badge in the first screen
  • add launch examples for Obsidian vaults
  • avoid presenting Kosmos as a replacement editor

2. Rebuild the first screen around trust

  • lead with "Can I trust this workspace?"
  • prioritize health score, missing sources, broken links, unsupported outputs, stale outputs, and recent AI activity
  • make "what needs review" more important than generic event volume

3. Make AI activity obvious

  • create a prominent "AI changed this" workflow
  • show agent-touched files by session
  • distinguish reads, writes, generated outputs, and prompt/instruction changes
  • alert when a session introduces broken links or unsupported outputs

4. Make provenance the core interaction

  • every output should show what it derives from
  • every note should show incoming and outgoing context evidence
  • every session should show the context path the agent traversed

5. Ship an Obsidian companion path

  • start with docs and CLI examples
  • later add an Obsidian plugin with commands like "Open in Kosmos" and "Run Context Audit"
  • keep the full graph, replay, and audit views in Kosmos instead of trying to rebuild them inside Obsidian

Product Test

Before adding a feature, ask:

Does this help a user understand what AI read, changed, cited, missed, or broke?

If yes, it belongs.

If no, it is probably a distraction until the core trust workflow is unmistakable.