Lorecrewpreview

Self-hosted AI teams

Build a crew.
Keep what
it learns.

For the code you ship. The ideas you explore. And all the useful experience worth keeping.

Named agents, durable work and shared knowledge in a Markdown vault. Run Lorecrew on your own infrastructure, with the models you choose.

Start your crew Explore the source
Open source · MIT Developer preview
Your knowledge
stays with you.✳
your crew / local workspace LOCAL
Make the next release feel right.Project workspace · owned here
Coding
Let’s make the settings page keyboard-friendly.
A

Ada Project lead

I’ll make a plan, work in the project checkout, and ask Lin to review the change.

L
Lin · review

Check focus order, labels and the regression check.

Useful experience

A lesson for the next interface.

Keep the keyboard check linked to the work that taught it.

Lesson candidate · evidence-linked
Illustrative workflow · no model calls on this website
One crew. A growing body of experience.↳ yours to inspect

A small stack.
A solid place to build.

Pi DurableMarkdownGitSQLiteNode.js

Make it your own

A team for your work.
Room for your whole day.

Start with a useful conversation. Add a project, a specialist or a routine when the work calls for it.

01 / BUILD

More capable project work.

Give a lead an objective, involve named specialists and keep tasks, questions and outcomes in context. Git worktrees and explicit ownership coordinate managed coding work.

Explore project workflows
02 / THINK

An assistant for everything around it.

Research, plan, draft and work with local files in a direct conversation. Bring your own notes and a model that suits the task—without setting up a coding project.

Meet your first agent

Beyond the current conversation

Useful work becomes
useful experience.

Keep decisions and lessons close to the work behind them. Search and reuse what matters, and inspect the evidence before calling it validated.

  1. 01

    Do the work

    Conversations and project tasks retain their identities, instructions and outcomes.

  2. 02

    Keep the lesson

    Capture source-linked experience in readable, portable Markdown.

  3. 03

    Check what holds up

    Distinguish tentative ideas from lessons supported by recorded checks.

  4. 04

    Bring it forward

    Later work can discover relevant knowledge and load exact skill revisions.

✳   Configure an optional background curator to help refine knowledge.Inside the learning loop

Your infrastructure. Your continuity.

Independent hosts.
A shared body of knowledge.

Run Lorecrew on a machine you control. Share accepted knowledge through Git while each instance keeps its own execution state and credentials.

  • Read the vault while every service is stopped.
  • Choose providers and models per agent.
  • Move project ownership through a deliberate stopped handoff.
Understand the architecture
Git-shared knowledge
Your Markdown vaultAgents · skills · lessons · history
lesson.md ↗
skill.md ↗
DesktopLocal instance
Home serverLocal instance
Another hostLocal instance
One active owner per project. Explicit handoff.
Remote views show synchronized history, not a live control channel.

Durable where it matters.

Resume local state after shutdown, reconcile original requests and inspect uncertain effects instead of guessing their outcome.

Recovery, deliberately

A calmer attention surface.

Run local routines and see meaningful results, questions and blockers in the Inbox. Keep quiet checks in run history.

Routines & Inbox

Your models. Clear choices.

Use the Pi AI provider catalog, supported local endpoints and model-specific thinking levels. Existing sessions keep their pins.

Models & thinking

Small start. Plenty of possibility.

Your next collaborator
starts here.

Node.js 24, npm 11 and Git. Build the developer preview, then create a private credential store, pair your browser and choose a vault.

Linux x64 is the currently verified execution platform. Read platform and security guidance before choosing another host.

Open the first-start guide
START FROM SOURCE
git clone https://github.com/Odiobill/lorecrew.git
cd lorecrew
npm ci --ignore-scripts --no-audit --no-fund
npm run build
MIT licensed · developer previewView repository ↗

A few good questions

Know what you’re
bringing home.

Can I use Lorecrew for more than coding?

Yes. Ordinary conversations support planning, research, drafting, knowledge retrieval and host files/tools without a coding project. Team delegation currently belongs to an owned project.

Does every model run locally?

You choose. Lorecrew uses the provider/model catalog from Pi AI, including supported local endpoints. A remote provider receives the model request you send to it. Self-hosting the service and running a local model are separate choices.

What stays in my vault?

Agent definitions, knowledge, memories, skills, lessons and readable history live in Git-backed Markdown. Private databases keep local execution and recovery records; credentials stay private to the host. Back up those separately.

How do multiple machines work?

Each instance runs independently and shares accepted knowledge through Git. Different hosts can work on different projects. A project has one confirmed owner, and a deliberate stopped handoff moves future execution.

What should I know before trying it?

This is a developer preview for a trusted single owner. Agents use the host account’s file and shell permissions. Linux x64 is the currently verified execution platform; other hosts need additional validation. Start with disposable work and read the security guide.

What about MCP and other integrations?

MCP, email/calendar connectors, push notifications and other extensions are future work. The current preview focuses on local agents, coding teamwork, useful knowledge, routines and deliberate sharing.

Open source. Built to be yours.

Good work should
leave something useful behind.

Make your first crew

Find your way

Browse all guidesEsc to close