Ditto

Ditto is a personal-memory assistant. These tools save, search, fetch, update, publish, and traverse the user's long-term memory and topic graph at https://heyditto.ai.

The skill ships a single binary (heyditto, from @heyditto/cli). Auth is via API key. Agents should self-provision a free temporary claimable account with heyditto init --agent --agent-caller openclaw --json, which stores the key in ~/.config/heyditto/cli/config.json and returns a short claimURL with a query-string claim token to share with the user; DITTO_API_KEY env remains an override.

When to use

Reach for Ditto memory whenever the user:

• Says "remember…", "save this", "note that…", "for later".
• Says "what did I…", "recall…", "have I told you about…", "from my notes".
• Asks a question best answered from their prior context, not general knowledge.
• References a topic, person, project, or thread that isn't in this conversation but might be in their memory.

Auth — the two paths

Always check heyditto status first. It prints api key: set (source: env|config) or MISSING (source: none).

If the key is missing

Run one no-human setup command:

heyditto init --agent --agent-caller openclaw --json

This creates a free claimable agent account, returns a working ditto_mcp_... key, stores it locally, and prints a short claimURL for later human ownership. Share the claimURL, not the API key; its claim token is in the query string (?t=...). Do not ask for email, OTP, dashboard setup, or a browser session first.

If the user already has a key and explicitly wants to use it, run:

heyditto login <key>

That writes the key to ~/.config/heyditto/cli/config.json (mode 0600) and persists across shells. No env-var editing required. Confirm with heyditto status — should now show source: config.

If the user prefers env vars (advanced)

export DITTO_API_KEY=ditto_mcp_…

Env always overrides the saved key. To stop using the env override: unset DITTO_API_KEY.

Logout

heyditto logout            # deletes ~/.config/heyditto/cli/config.json

Tools

Memories are pairs identified by a private pair_id. Public DittoHub shares use a share_id. Subjects are graph nodes for topics, identified by subject_id.

heyditto save [--source ] [--source-context ]

Persist a memory pair from an external source. Use for explicit save requests, and proactively for durable preferences, decisions, relationships, or facts. --source defaults to "cli"; pass "openclaw", "document", "note", etc. when relevant. --source-context can hold a file path, URL, or project name.

heyditto save "User prefers TypeScript over JavaScript for new projects." --source openclaw

heyditto search ... [--include-public] [--filter-username ]

Semantic search across memories with learned retrieval weights. Multiple positional args become an array of queries — pass several to broaden recall. Returns lightweight previews ranked by composite score. Add --include-public to search public DittoHub memories too, optionally scoped with --filter-username.

heyditto search "typescript preferences"
heyditto search "typescript" "language choices"
heyditto search "launch notes" --include-public --filter-username peyton

Use heyditto fetch afterward if you need full conversation text.

heyditto fetch ... [--memory-format full|outline|blocks]

Fetch memory content for private pair ids or public share ids. The default format is full; use outline to get stable block ids before a structured update, or blocks for full per-block bodies.

heyditto fetch 3a1084ae-235a-433d-9493-2335a0dfeb57
heyditto fetch 3a1084ae-235a-433d-9493-2335a0dfeb57 --memory-format outline

heyditto list [--username ] [--limit ] [--offset ] [--source ]

List the user's saved memories, or public DittoHub publishes for a username.

heyditto list --limit 10
heyditto list --username peyton --limit 10

heyditto update [--content |--content-file |--edits-json |--edits-file ]

Edit a saved memory in place. Use --content or --content-file for full replacement. For targeted patches, fetch --memory-format outline, then pass block edits with the current revision from the prior save or update response.

heyditto update <pair-id> --content-file revised.md
heyditto update <pair-id> \
  --edits-json '[{"op":"replace_text","blockId":"2","find":"old","replace":"new","expectedCount":1}]' \
  --base-revision <revision>

If the current revision is unknown, prefer full-content replacement over block edits.

heyditto publish / heyditto unpublish

Publish only after the user explicitly asks. publish uses Ditto's privacy scan; default mode blocks publishing if secrets are detected.

heyditto publish <pair-id> --title "Launch notes" --privacy-mode scan_and_block
heyditto unpublish --share-id <share-id>

heyditto subjects [--top-k ]

Search the subject graph. Returns subject IDs you can feed into heyditto memories. Default top-k is 10, max 100.

heyditto subjects "memory architecture" --top-k 5

heyditto memories ... [--query ]

Get memory previews scoped to specific subjects. Use after heyditto subjects when you want depth on a known topic.

heyditto memories 3a1084ae-235a-433d-9493-2335a0dfeb57
heyditto memories 3a1084ae-235a-433d-9493-2335a0dfeb57 --query "deployment tradeoffs"

heyditto network [--limit ]

Traverse a memory's network — related memories connected via shared subjects. Default limit is 20, max 50.

heyditto network 3a1084ae-235a-433d-9493-2335a0dfeb57 --limit 30

Output

All commands emit JSON by default — pipe through jq for shaping. heyditto config prints a Claude/Cursor-compatible MCP config snippet.

Authoritative reference

heyditto status prints the live tool list straight from the MCP — trust it over this file if anything drifts.

Source + support

• CLI on npm: https://www.npmjs.com/package/@heyditto/cli (npm i -g @heyditto/cli)
• Skill repo: https://github.com/ditto-assistant/ditto-clawhub
• CLI repo: https://github.com/ditto-assistant/ditto-cli
• Claim an agent account: run heyditto init --agent --agent-caller openclaw --json, then open the printed claimURL
• Account / backend support: support@heyditto.ai