Cloud knowledge backup and retrieval using Supermemory.ai free tier. Store high-value insights to the cloud and search them back when local memory is insufficient. Uses standard /v3/documents and /v3/search endpoints (no Pro-only features).
Backs up important knowledge and insights to Supermemory.ai's cloud using the free tier API.
Uses only /v3/documents (store) and /v3/search (retrieve) — no Pro-only endpoints.
Set in .env
SUPERMEMORY_OPENCLAW_API_KEY="sm_..."
Store a knowledge string to the cloud.
python3 skills/supermemory-free/store.py "Your knowledge string here"
# With optional container tag (namespace/filter)
python3 skills/supermemory-free/store.py "knowledge string" --tag openclaw
# With metadata
python3 skills/supermemory-free/store.py "knowledge string" --tag fixes --source "session"
# Output raw JSON
python3 skills/supermemory-free/store.py "knowledge string" --json
When to use:
Search the cloud memory for relevant knowledge.
python3 skills/supermemory-free/search.py "your query"
# With container tag filter
python3 skills/supermemory-free/search.py "your query" --tag openclaw
# More results
python3 skills/supermemory-free/search.py "your query" --limit 10
# Higher precision (less noise)
python3 skills/supermemory-free/search.py "your query" --threshold 0.7
# Search across ALL tags
python3 skills/supermemory-free/search.py "your query" --no-tag
When to use:
Scans recent session memory logs and automatically pushes high-value insights to Supermemory cloud.
# Run manually
python3 skills/supermemory-free/auto_capture.py
# Dry run (show what would be captured, no upload)
python3 skills/supermemory-free/auto_capture.py --dry-run
# Scan last N days (default: 3)
python3 skills/supermemory-free/auto_capture.py --days 7
# Force re-upload even if already seen
python3 skills/supermemory-free/auto_capture.py --force
# Verbose mode
python3 skills/supermemory-free/auto_capture.py --verbose
Install cron job (runs daily at 2:00 AM UTC):
bash skills/supermemory-free/install_cron.sh
Remove cron job:
bash skills/supermemory-free/install_cron.sh --remove
Check cron status:
bash skills/supermemory-free/install_cron.sh --status
The auto-capture script identifies "high-value" insights from memory logs using these heuristics:
| Pattern | Label | Example |
|---|---|---|
| Resolved errors / fixes | fix | Fixed: SSL cert error by running... |
| Error context | error | Exception: Connection refused on port 5432 |
| Configuration paths | config | /etc/nginx/sites-available/default |
| API/endpoint info | api | Endpoint: POST /v3/documents for storage |
| User preferences | preference | User prefers Python over Node for scripts |
| Decisions made | decision | Decided to use PostgreSQL because... |
| Learned facts | insight | Learned that cron syntax for... |
| Installs / setup | setup | Installed nginx, configured with... |
| Bullet-point blocks | bullet | - Key finding: X works better than Y |
Deduplication: Already-uploaded items are tracked in .capture_state.json — re-running is safe.
Use --tag to namespace your memories:
| Tag | Purpose |
|---|---|
openclaw | General OpenClaw session knowledge (default) |
fixes | Bug fixes and solutions |
config | Configuration and setup |
user-prefs | User preferences |
projects | Project-specific knowledge |
| File | Purpose |
|---|---|
store.py | CLI tool: upload knowledge to cloud |
search.py | CLI tool: search cloud knowledge |
auto_capture.py | Cron script: auto-analyze memory logs |
install_cron.sh | Install/remove/status of cron job |
.capture_state.json | Dedup state (auto-generated, gitignore) |
SKILL.md | This file |
_meta.json | Skill metadata |
https://api.supermemory.aiPOST /v3/documentsPOST /v3/searchSUPERMEMORY_OPENCLAW_API_KEYHTTP 403 / 1010 Access Denied:
The scripts include proper User-Agent, Origin, and Referer headers to satisfy Cloudflare. If it recurs, verify the API key is valid at https://console.supermemory.ai.
No memory files found:
Auto-capture looks in memory/YYYY-MM-DD.md. Ensure your memory skill is writing daily logs there.
Re-upload everything:
Delete .capture_state.json or use --force to ignore the dedup state.