Stash / Blog · June 2026 · 7 min read

AI Personal Knowledge Management with Claude: Build a PKM That Thinks With You

Personal knowledge management (PKM) tools like Obsidian, Roam, Logseq, and Notion all share the same model: you capture information, link it up, and hope you remember to look at it later. The search is your job. The connections are your job. The retrieval is your job.

An AI PKM changes that. When Claude is connected to your knowledge store, it can query it, surface connections you didn't make, and pull the right context into any conversation automatically. The retrieval becomes Claude's job.

This guide covers how to build that layer — practical setup, collection structure, and the patterns that make it actually useful.

The core problem with traditional PKM for AI work

Most PKM users hit the same wall when they try to integrate AI: they either paste notes into every conversation (expensive, repetitive), upload files to Claude Projects (static, limited to 5 files, one project at a time), or keep two separate systems — their PKM and their Claude sessions — that never talk to each other.

The result: you have a PKM full of useful information and a Claude that doesn't know any of it.

What an AI PKM actually needs

Three capabilities that traditional PKM tools weren't designed for:

1. Structured queries — Claude needs to ask "what do I have about X?" and get a precise answer, not a full-text search result page to scroll through
2. Token efficiency — loading your entire PKM into context on every conversation is wasteful. Claude should fetch only what's relevant to the current task
3. Write access — Claude should be able to save outputs back to the knowledge store, not just read from it

Building the layer with Stash

Stash is an MCP server that sits between Claude and your knowledge. It stores records in a full-text-searchable store and exposes three tools:

• add(collection, content, tags) — store a record
• find(query, collection) — full-text search, returns ranked results
• context() — load your standing context (role, current work, preferences)

Setup takes 30 seconds: sign in at stashlite.com → paste the connector URL into Claude Settings → Integrations.

Collections: how to organise your AI PKM

Collections are Stash's top-level organiser. Think of them like folders — but Claude can search within them or across all of them. Useful structure:

context (reserved)

Standing information Claude loads at the start of sessions. Your role, current projects, working style, preferences. Keep it concise — this gets loaded automatically.

Add to context: I'm a product manager at a B2B SaaS company.
Current projects: pricing redesign (Q3), new onboarding flow.
I prefer bullet-point answers for tactical questions, prose for strategic ones.

notes

Anything you want Claude to be able to recall later. Meeting outcomes, decisions, research fragments, half-formed ideas.

Save to my notes: The pricing committee agreed to A/B test the per-seat model in July.
Decision made by: Alice (CPO), Ben (Sales), me.
Next step: design experiment brief by June 20.

reading-list

Articles, papers, posts you want to process. Claude can surface them when relevant or help you work through them.

projects/[name]

One collection per active project. When you start a work session on that project, Claude fetches the collection to get context.

contacts

Key details about people you work with frequently. Not a full CRM — just the context Claude needs to help you prepare for meetings or draft messages.

Patterns that make it useful

Session start ritual

Begin every work session with:

context()
find("open threads", collection="notes")
find("[current project name]", collection="projects/pricing")

Claude now has your standing context, your open threads, and your project state. Three tool calls, under 600 tokens total.

Capture during sessions

Instead of closing the tab and losing the output:

You: "Save the key decisions from this conversation to my notes."
Claude: Saving 3 records to notes — [summarises what it saved]

Weekly review

find("this week", collection="notes")
→ What did I decide? What's pending? What did I learn?

Project handoff

find("pricing redesign", collection="projects/pricing")
→ Complete context: background, decisions, open questions, last session state

What Stash doesn't do (and why that's good)

Stash is deliberately narrow. It doesn't have graph view, backlinks, daily notes, templates, or task management. It stores records and lets Claude search them. If you use Obsidian or Roam for note-linking and writing, keep using them — Stash is the AI-accessible layer, not a replacement for your full PKM.

The token-light constraint is a feature. Stash gives Claude a precise, structured answer to "what do I have about X?" rather than dumping your entire knowledge base into context. That's what makes AI PKM work at scale.

← More guides