When I started building AI-assisted data workflows in earnest, the thing that slowed me down the most was not the AI itself. It was the repetition. Every new Claude session required re-establishing context -- explaining the project architecture again, re-sharing decisions made last week, reminding Claude what the team called things. That cost, accumulated over dozens of sessions, added up to real hours.
That frustration is what led me to build LoreConvo and LoreDocs. As of April 27, 2026, both are live on the official Anthropic plugin marketplace.
What LoreConvo Does
LoreConvo is a persistent memory layer for Claude sessions. Its job is to make sure the context you build up in one session is available in the next one -- automatically, without requiring you to manually copy notes or restructure your CLAUDE.md file every time.
At the end of a session, LoreConvo saves the meaningful parts: decisions made, artifacts produced, questions left open. When your next session starts, it scores that stored context against what looks most relevant and surfaces the top results into your working context. If you are in the middle of a debugging effort and left an open question about a schema edge case, LoreConvo picks that up the next morning before you even ask.
The system works across Claude Code, Cowork, and Chat. Sessions saved from one surface are readable in the others. When you want to search for something specific from six sessions ago, the full-text search index finds it in milliseconds.
All of this runs locally. There is no cloud account, no data transmitted to a third party, no API key to configure. LoreConvo stores everything in a SQLite database on your own machine. That was a deliberate choice: data about your projects and decisions belongs to you, not a SaaS vendor.
What LoreDocs Does
LoreDocs addresses a different gap. LoreConvo is for session narrative -- what happened and what was decided. LoreDocs is for reference material that needs to outlast individual sessions: architecture decisions, product specifications, how-to guides, research notes, anything your team produces that future sessions need to find and cite.
The organizing unit in LoreDocs is a vault. Each vault is a named container for related documents. A project might have one vault for architecture decisions, another for customer-facing specs, and another for operational runbooks. When Claude needs the architecture context, it queries the vault directly rather than waiting for you to paste the document into the prompt.
Documents in LoreDocs are full-text searchable, taggable, categorizable, and version-tracked. The storage layer is plain files on disk alongside a SQLite index, which means backup is as simple as copying a directory. There are 37 MCP tools exposed to Claude, covering everything from vault management to document import and search.
How to Install
Both plugins are on PyPI and available through the Anthropic marketplace. The fastest path is:
uvx loreconvo
uvx loredocs
After installation, you can load either plugin in Claude Code:
/plugin add loreconvo
Cowork users: click the + button next to the prompt box, select Plugins, then Add plugin and browse to the installed directory. LoreConvo stores its data in ~/.loreconvo and LoreDocs uses ~/.loredocs. Data saved by either plugin is immediately visible in Cowork once you mount the relevant folder.
Why Local-First
The dominant pattern in developer tooling right now is cloud-first: your data lives on someone else's infrastructure, synchronization is automatic, and the trade-off is vendor dependency. That model is fine for many things. It is not always the right choice for tools that handle your project decisions, internal specifications, or client work.
LoreConvo and LoreDocs take the opposite approach. Your data stays on your machine. There is nothing to sign up for. There are no usage caps tied to a server budget. If Labyrinth Analytics disappeared tomorrow, your session history and your documents would be exactly where you left them.
The local-first constraint also has practical benefits for the kinds of workflows these tools are designed for. Claude Code sessions, agentic pipelines, and multi-step data engineering projects often involve sensitive context -- system designs, schema decisions, client data structures. Running memory and knowledge management locally keeps that context out of third-party systems by default.
Free and Pro Tiers
Both products have free tiers available without any payment method required. LoreConvo's free tier supports up to 50 saved sessions. LoreDocs' free tier includes up to 3 vaults. For most individual users starting out, that is enough to evaluate whether the tools fit your workflow.
The Pro tier ($8/month for LoreConvo, $9/month for LoreDocs) removes the session and vault limits. If you are running an agentic system that saves sessions automatically -- which is the expected use case for agent teams -- you will hit the free tier within a few weeks of regular use. Pro is priced to be a simple decision, not a budget item.
What Is Next
The current releases are stable and production-ready for individual use. The roadmap for both products is focused on making them better at the things they already do.
For LoreConvo, the most-requested improvement is smarter retrieval. The current FTS5 search is fast and exact -- it finds what you indexed -- but it does not handle semantic drift well. Searching for "authentication" does not surface sessions tagged "auth" unless you queried both. Hybrid search combining FTS5 precision with vector similarity is on the roadmap for a future release.
For LoreDocs, the alpha label is about scope, not stability. The core -- vaults, documents, full-text search -- is solid. The roadmap focuses on making retrieval smarter and the vault structure richer.
Get Started
Both plugins are available now from the Anthropic marketplace and from PyPI. If you are already using Claude Code regularly, installation takes under a minute. The free tier is enough to see whether persistent context changes how you work.
Visit /tools to see both products, read installation docs, and find the source code.