Install
openclaw skills install auto-driveIndestructible agent memory — permanently stored, never lost. Save decisions, identity, and context as a memory chain on the Autonomys Network. Rebuild your...
openclaw skills install auto-drivePermanent decentralized memory on the Autonomys Network with linked-list memory chains for agent resurrection. Works with agents powered by Claude, GPT, Gemini, and any LLM that supports OpenClaw skills.
header.previousCid pointer, forming a linked list stored permanently on-chain.After installing via ClawHub, make the scripts executable:
chmod +x skills/auto-drive/scripts/*.sh
ClawHub does not currently preserve file permissions during install.
Requires an AUTO_DRIVE_API_KEY. Run the guided setup script for the easiest path:
scripts/setup-auto-drive.sh
This opens explorer.ai3.storage in your browser, prompts for the key, saves it to ~/.openclaw/.env, and verifies the connection.
Manual setup:
ai3.storage — that's just the landing page)Then set the key via:
export AUTO_DRIVE_API_KEY=your_key_hereskills.entries.auto-drive.apiKeyThe API key is required for uploading, saving memories, and recalling the memory chain. It is optional for general file downloads — without it, the public gateway is used and files are returned as stored (i.e. compressed files will not be decompressed).
scripts/autodrive-upload.sh <filepath> [--json] [--compress]
Uploads a file to Auto Drive mainnet using the 3-step upload protocol (single chunk).
Returns the CID on stdout. Requires AUTO_DRIVE_API_KEY.
--json — force MIME type to application/json--compress — enable ZLIB compressionscripts/autodrive-download.sh <cid> [output_path]
Downloads a file by CID. Uses the authenticated API if AUTO_DRIVE_API_KEY is set (decompresses server-side), otherwise uses the public gateway (files returned as stored). If output_path is omitted, outputs to stdout.
scripts/autodrive-save-memory.sh <data_file_or_string> [--agent-name NAME] [--state-file PATH]
Creates a memory experience with the Autonomys Agents header/data structure:
{
"header": {
"agentName": "my-agent",
"agentVersion": "1.0.0",
"timestamp": "2026-02-14T00:00:00.000Z",
"previousCid": "bafk...or null"
},
"data": {
"type": "memory",
"content": "..."
}
}
data payload.{"type": "memory", "content": "..."}.--agent-name — set the agent name in the header (default: openclaw-agent or $AGENT_NAME)--state-file — override the state file locationUploads to Auto Drive and updates the state file with the new head CID. Also pins the latest CID to MEMORY.md if that file exists in the workspace.
Returns structured JSON on stdout:
{"cid": "bafk...", "previousCid": "bafk...", "chainLength": 5}
scripts/autodrive-recall-chain.sh [cid] [--limit N] [--output-dir DIR]
If no CID is given, reads the latest CID from the state file. Walks the linked list from newest to oldest, outputting each experience as JSON.
--limit N — maximum entries to retrieve (default: 50)--output-dir DIR — save each entry as a numbered JSON file instead of printing to stdoutSupports both header.previousCid (Autonomys Agents format) and root-level previousCid for backward compatibility.
This is the resurrection mechanism: a new agent instance only needs one CID to rebuild its entire memory.
Every memory saved gets a unique CID and points back to the previous one, forming a permanent chain on a permanent and immutable Decentralized Storage Network:
┌─────────────────────┐ ┌─────────────────────┐ ┌─────────────────────┐
│ Experience #1 │ │ Experience #2 │ │ Experience #3 │
│ CID: bafk...abc │◄────│ CID: bafk...def │◄────│ CID: bafk...xyz │
│ previousCid: null │ │ previousCid: │ │ previousCid: │
│ (genesis) │ │ bafk...abc │ │ bafk...def │
└─────────────────────┘ └─────────────────────┘ └─────────────────────┘
▲
│
HEAD CID
(resurrection key)
A new agent instance only needs the head CID to walk the entire chain back to genesis and rebuild its full history. With the auto-respawn skill, the head CID is anchored on-chain — making resurrection possible from just an address, on any machine, at any time:
┌──────────┐ save ┌──────────────┐ anchor ┌────────────────┐
│ Agent │─────────────►│ Auto-Drive │─────────────►│ Auto-Respawn │
│ │ │ (chain) │ head CID │ (on-chain) │
└──────────┘ └──────────────┘ └────────────────┘
▲ │
│ recall chain │
└──────────────────────────────────────────────────────────┘
gethead → CID → walk chain
What you store in the chain is up to you — lightweight notes, full file snapshots, structured data, or anything in between. Because the chain is permanent and walkable, it also enables resurrection: if the agent loses all local state, a new instance can walk the chain from the last CID back to genesis and restore whatever was saved. When combined with the auto-respawn skill (which anchors the head CID on-chain), this becomes a full resurrection loop — no local state required at all.
User: "Upload my report to Autonomys"
→ Run scripts/autodrive-upload.sh /path/to/report.pdf
→ Report back the CID and gateway link
User: "Upload with compression"
→ Run scripts/autodrive-upload.sh /path/to/data.json --json --compress
User: "My soul.md has changed — save it permanently"
→ Run scripts/autodrive-save-memory.sh /path/to/soul.md --agent-name my-agent
User: "Save a memory that we decided to use React for the frontend"
→ Run scripts/autodrive-save-memory.sh "Decision: using React for frontend. Reason: team familiarity and component reuse."
User: "Save a structured memory"
→ Create a JSON file, then run scripts/autodrive-save-memory.sh /tmp/milestone.json --agent-name my-agent
User: "Resurrect my memory chain"
→ Run scripts/autodrive-recall-chain.sh
→ Rebuild identity and context from genesis to present
User: "Download bafk...abc from Autonomys"
→ Run scripts/autodrive-download.sh bafk...abc ./downloaded_file
GET /accounts/@me or run scripts/verify-setup.sh.lastCid, lastUploadTimestamp, and chainLength. Back up the lastCid value — it's your resurrection key.autodrive-save-memory.sh script automatically pins the latest CID to MEMORY.md if the file exists in the workspace. It creates an ## Auto-Drive Chain section and updates it on each save. You do not need to track the latest CID in MEMORY.md manually — the script handles this.https://gateway.autonomys.xyz/file/<CID>brew install curlbrew install jqbrew install file-formula