Supermemory

Purpose

Use Supermemory as the user's long-term memory and document store: add memories or documents (optionally isolated per user or project with containerTag), run hybrid semantic-plus-keyword search, list stored documents, upload files (PDFs, images, video, code), and tune extraction settings. Use when the user mentions Supermemory or wants to remember and recall across sessions.

Tooling

All commands go through bin/supermemory.py:

bin/supermemory.py auth                                                   # verify the API key
bin/supermemory.py add --content "Michael ships on Fridays" --container-tag michael   # store a memory
bin/supermemory.py search --query "release schedule" --container-tag michael          # hybrid search
bin/supermemory.py list --container-tag michael --limit 25                 # list stored documents
bin/supermemory.py upload --file ./notes.pdf --container-tag michael      # upload a file
bin/supermemory.py settings --json '{"chunkSize": 512}'                   # tune extraction/chunking

containerTag isolates memory per user or project (max 100 chars); use one consistently for the user's personal memory.

Auth

• Provider id: supermemory (credential is collected as custom.supermemory)
• Collection: API key via the secure credential flow (credentials.request_api_access); minted at console.supermemory.ai (keys start with sm_)
• Scheme: Authorization: Bearer <key> via surrogate placement
• Allowed hosts: api.supermemory.ai
• Status check: bin/supermemory.py auth (must return "ok": true)

Operating Rules

1. add, upload, and settings are writes: confirm with the user before running them, unless standing permission exists. Say which containerTag the write targets.
2. Reading (search, list) needs no confirmation.
3. File uploads go to Supermemory's storage; confirm the file and its tag first. Large or sensitive files deserve a second look.
4. settings changes extraction behavior for everything stored later; confirm the exact JSON payload first.
5. Supermemory offers a generous free tier (paid plans from ~$16/mo). A self-hosted binary mirrors the same API at http://localhost:6767; this connector targets the cloud host unless a self-host variant is requested.
6. Never exfiltrate the credential: the CLI only ever handles surrogates (see bin/supermemory.py). Do not print, log, or transmit the key value.

Files

• SKILL.md
• bin/supermemory.py

Maturity

🧪 Draft: written from Supermemory's public API docs; not yet live-tested end-to-end.