Import a Claude project

Share the instructions and memory Claude Code keeps for a team project with everyone on it, and keep the copy current with one command.

Before you begin

A lore login completed against your server. See Import knowledge.
A project you have used with Claude Code, so it has a CLAUDE.md or a memory directory under ~/.claude/projects/.
A role in lore.json that the project’s teammates hold, such as payments-team.
Understand how the knowledge gateway turns a synced store into a docset.

1. Connect the project

Change into the project directory, the one you run Claude Code in, and run the guided import from there. lore finds your knowledge repository through your last lore login, so you can run this from any folder. lore agent import claude and lore source add claude run the same operation.

lore agent import claude --project

--project offers only the groups that belong to this project and ticks the two Claude Code fills on its own: the instructions in the project folder and the auto memory Claude Code keeps for it under ~/.claude/projects/, in a directory named after the project path. lore works that directory out for you:

Connecting a synced Claude source, not a one-off import.
┃ Choose groups to sync
┃ > [•] Project instructions · ~/src/payments · 3 files
┃   [•] Project auto memory · ~/.claude/projects/-Users-alice-src-payments/memory · 1 file
┃   [ ] Session transcripts · ~/.claude/projects/-Users-alice-src-payments · 1 file

x toggle • ↑ up • ↓ down • / filter • enter submit • ctrl+a select all

Press Enter to keep the ticked groups. lore lists every file in each one and asks once before writing your sources.yaml:

payments-claude-instructions → claude/alice/-Users-alice-src-payments/instructions (3 files)
    .claude/rules/testing.md
    AGENTS.md
    CLAUDE.md

payments-claude-memory → claude/alice/-Users-alice-src-payments/memory (1 file)
    MEMORY.md

┃ Write these synced source definitions to sources.yaml?
┃
┃                      Yes     No

←/→ toggle • enter submit • y Yes • n No

Press y. Each group becomes one source in home/<identity>/sources.yaml, where <identity> is the name you logged in with. Pass a path after claude to import a project other than the current directory, --name to replace the payments-claude prefix, and --dry-run to preview without writing. Esc cancels either prompt and connects nothing.

The groups --project offers, and where each syncs to:

Project instructions: CLAUDE.md, .claude/CLAUDE.md, .claude/rules/**, CLAUDE.local.md and AGENTS.md in the project. Store claude/<identity>/<project-key>/instructions.
Project auto memory: everything under ~/.claude/projects/<project-key>/memory/. Store claude/<identity>/<project-key>/memory.
Session transcripts: the *.jsonl conversation logs under ~/.claude/projects/<project-key>/. Store claude/<identity>/<project-key>/transcripts. Never ticked by default; press x on it to include them.

<project-key> is Claude Code’s own name for the project directory: the absolute path with every non-alphanumeric character replaced by -. Without --project, the picker also offers Personal instructions from ~/.claude, which Import your Claude memory covers.

Note: Settings, agent memory, .env files and any file whose name looks like a credential, key or token are always excluded, even with a broad --include. Transcripts are not: they are off by default because they record whole conversations, and nothing filters their contents, so read them before you tick the group. Filename filtering does not read file contents, so check the list before confirming.

2. Sync

Push the chosen groups to the server:

lore sync

You should see:

Syncing as alice to https://lore.example.com. Identity-scoped sources:
  payments-claude-instructions → claude/alice/-Users-alice-src-payments/instructions
  payments-claude-memory → claude/alice/-Users-alice-src-payments/memory
┃ Push these sources under this identity?
┃
┃                      Yes     No
←/→ toggle • enter submit • y Yes • n No

Press y. The question appears once per server and identity on this machine, and each source then reports what it uploaded:

payments-claude-instructions: 3 files · 3 new blobs · +3 ~0 -0 · snapshot 9b1e4c70…
payments-claude-memory: 1 files · 1 new blobs · +1 ~0 -0 · snapshot c47d02a1…

Run lore sync again whenever Claude Code updates its memory; only new content is uploaded.

3. Give the team access

Synced content is invisible until a docset projects it. Add one projection per store to lore.json and grant the roles your teammates hold, exactly as for any other docset. Edit the file yourself, or ask an agent that already has the knowledge repository open to do it:

lore.json
Ask your agent

{
  "docsets": {
    "payments-claude": {
      "projection": {
        "store": "claude/alice/-Users-alice-src-payments/instructions"
      },
      "access": { "allow": { "payments-team": "ro" } }
    },
    "payments-memory": {
      "projection": {
        "store": "claude/alice/-Users-alice-src-payments/memory",
        "include": ["**/*.md"]
      },
      "access": { "allow": { "payments-team": "ro", "platform": "ro" } }
    }
  }
}

Run `lore source list` and find the two stores for the payments Claude import. In lore.json, add a docset named payments-claude that projects the instructions store and a docset named payments-memory that projects the memory store with only **/*.md included. Grant payments-team read-only access to both and platform read-only access to payments-memory. Keep every other docset unchanged.

Run lore config reload in an OpenLore session with lore:config:edit; no server restart is needed. The docsets mount read-only at /payments-claude and /payments-memory, including in existing sessions, and everyone who holds payments-team reads both; platform reads only the memory. Add a role to access.allow to widen the audience and remove it to narrow it, then reload again; a projection never accepts writes, whatever grant you give.

4. Commit the definitions

home/<identity>/sources.yaml now holds the two sources with a portable project path:

version: "1"
sources:
  - name: payments-claude-instructions
    type: claude
    group: instructions
    project: ~/src/payments
  - name: payments-claude-memory
    type: claude
    group: memory
    project: ~/src/payments

Commit it with lore.json to the knowledge repository. A teammate who wants to contribute their own Claude Code files for the same project runs the same import under their own identity; their sources land in their own home/<identity>/sources.yaml and their own claude/<identity>/… stores, and a second projection makes those readable too.

Verify

Read the synced memory through the server as a teammate who holds payments-team. Set OPENLORE_SERVER to your server’s hostname first, for example lore.example.com, or localhost for a local server:

ssh -p 2222 "$OPENLORE_SERVER" "cat /payments-memory/MEMORY.md"