---

name: dht description: "Routes Hivemind DHT workflows for starting peers, storing and fetching records, and debugging multiaddress connectivity." disable-model-invocation: true metadata: disco-role: operating license: MIT

Hivemind DHT

Use this route when the task is about hivemind.DHT, hivemind.dht.DHTNode, peer discovery, record storage, bootstrap peers, relay settings, or the hivemind-dht command.

Include

• Creating a first DHT node or joining an existing swarm.
• Storing and fetching DHT records, including subkey records and expiring values.
• Custom run_coroutine(...) logic that runs inside the DHT process.
• Validators, signed records, peer identity, and bootstrap/connectivity debugging.
• Multiaddress choices for host_maddrs, announce_maddrs, initial_peers, client_mode, use_relay, use_auto_relay, and use_ipfs.
• The hivemind-dht console command.

Exclude

• Collaborative averaging, optimizers, gradient/state sharing, and compression strategy selection; route those to collaborative-training.
• Hosted experts, remote expert clients, and MoE routing; route those to moe.
• Benchmarks and CI-only connectivity experiments.

Start here

1. Read references/api-reference.md for the verified signatures and key options.
2. Read references/workflows.md for step-by-step DHT setup and store/get examples.
3. Read references/troubleshooting.md when bootstrap, relay, or identity errors appear.
4. Run ../../scripts/check_install.py from the generated skill tree to confirm the installed package and CLI entry points.

What to remember

• hivemind.DHT(start=True) starts a background DHT process.
• The first peer normally exposes local-loopback multiaddrs; add host_maddrs and announce_maddrs for externally reachable peers.
• store(..., expiration_time=...) and get(..., latest=...) are the common record operations.
• Use run_coroutine(...) for custom DHT-side logic, but do not call the external DHT interface from inside the DHT process itself.
• For restrictive firewalls or NAT, prefer client_mode=True, relay options, or use_ipfs=True depending on the deployment.

Good follow-up questions

• "How do I start a DHT node and print the peer addresses?"
• "How do I connect two Hivemind peers?"
• "Why does DHT startup fail with my bootstrap peers?"
• "How do I store and read a record from the DHT?"
• "How do I make a peer reachable across the internet?"