diff --git a/internal/stacks/ipfs_config.go b/internal/stacks/ipfs_config.go
index e05f854..1777d65 100644
--- a/internal/stacks/ipfs_config.go
+++ b/internal/stacks/ipfs_config.go
@@ -34,6 +34,14 @@ func GenerateSwarmKey() (string, error) {
 // disables Kubo's AutoConf/public-network defaults, which are incompatible
 // with a private swarm (swarm.key / LIBP2P_FORCE_PNET) and otherwise prevent
 // the daemon from starting or from ever peering with other members.
+//
+// Routing.Type is set to "none" rather than "dht": a small private swarm hits
+// a long-standing upstream bug (https://github.com/ipfs/kubo/issues/8346)
+// where bitswap finds a provider via the DHT but never sends it a WANT
+// message if the swarm connection to that peer already existed, so content
+// fetches hang even though the peers are connected. With routing disabled,
+// bitswap has no choice but to broadcast wants directly to its connected
+// peers, which is all a 2+ node private swarm actually needs.
 func GenerateIPFSPrivateNetInitScript() string {
 	return `#!/bin/sh
 ipfs config --json AutoConf.Enabled false
@@ -41,7 +49,7 @@ ipfs config --json Bootstrap '[]'
 ipfs config --json DNS.Resolvers '{}'
 ipfs config --json Routing.DelegatedRouters '[]'
 ipfs config --json Ipns.DelegatedPublishers '[]'
-ipfs config Routing.Type dht
+ipfs config Routing.Type none
 ipfs config --json AutoTLS.Enabled false
 ipfs config --json Swarm.Transports.Network.Websocket false
 `
diff --git a/internal/stacks/stack_manager.go b/internal/stacks/stack_manager.go
index d95fd8e..5c55c46 100644
--- a/internal/stacks/stack_manager.go
+++ b/internal/stacks/stack_manager.go
@@ -22,6 +22,7 @@ import (
 	"fmt"
 	"net"
 	"net/http"
+	"net/url"
 	"os"
 	"os/exec"
 	"path"
@@ -600,6 +601,8 @@ func (s *StackManager) peerIPFSNodes() error {
 		return nil
 	}
 
+	s.Log.Info("peering IPFS nodes")
+
 	type ipfsIDResponse struct {
 		ID string `json:"ID"`
 	}
@@ -614,30 +617,55 @@ func (s *StackManager) peerIPFSNodes() error {
 		peerIDs[member.ID] = idResp.ID
 	}
 
+	type peeringPeer struct {
+		ID    string   `json:"ID"`
+		Addrs []string `json:"Addrs"`
+	}
+
 	for _, member := range s.Stack.Members {
+		peers := []peeringPeer{}
 		for _, other := range s.Stack.Members {
 			if member.ID == other.ID {
 				continue
 			}
-			addr := fmt.Sprintf("/dns4/ipfs_%s/tcp/4001/p2p/%s", other.ID, peerIDs[other.ID])
+			peers = append(peers, peeringPeer{
+				ID:    peerIDs[other.ID],
+				Addrs: []string{fmt.Sprintf("/dns4/ipfs_%s/tcp/4001", other.ID)},
+			})
+		}
+		peersJSON, err := json.Marshal(peers)
+		if err != nil {
+			return err
+		}
 
-			// swarm/peering/add only registers the peer with Kubo's background
-			// reconnect service - it returns success even when the peer is
-			// unreachable, so it does not confirm a connection was made.
-			peeringURL := fmt.Sprintf("http://127.0.0.1:%d/api/v0/swarm/peering/add?arg=%s", member.ExposedIPFSApiPort, addr)
-			if err := core.RequestWithRetry(s.ctx, http.MethodPost, peeringURL, nil, nil); err != nil {
-				return fmt.Errorf("failed to peer IPFS node %s with %s: %w", member.ID, other.ID, err)
-			}
+		// swarm/peering/add only registers the peer with the *running*
+		// daemon's in-memory peering service and returns success even when
+		// the peer is unreachable - it does not persist to Peering.Peers in
+		// the on-disk config, so the pairing is lost on the next restart.
+		// Writing through /api/v0/config persists it for real, and Kubo
+		// picks up the change live (no restart needed).
+		configURL := fmt.Sprintf("http://127.0.0.1:%d/api/v0/config?arg=Peering.Peers&arg=%s&json=true", member.ExposedIPFSApiPort, url.QueryEscape(string(peersJSON)))
+		if err := core.RequestWithRetry(s.ctx, http.MethodPost, configURL, nil, nil); err != nil {
+			return fmt.Errorf("failed to persist IPFS peering config for member %s: %w", member.ID, err)
+		}
 
+		for _, other := range s.Stack.Members {
+			if member.ID == other.ID {
+				continue
+			}
+			addr := fmt.Sprintf("/dns4/ipfs_%s/tcp/4001/p2p/%s", other.ID, peerIDs[other.ID])
 			// swarm/connect dials synchronously and errors if the connection
 			// fails, so retrying it confirms the nodes are actually connected
-			// rather than just registered to reconnect in the background.
+			// now rather than waiting on the peering service's own timing.
 			connectURL := fmt.Sprintf("http://127.0.0.1:%d/api/v0/swarm/connect?arg=%s", member.ExposedIPFSApiPort, addr)
 			if err := core.RequestWithRetry(s.ctx, http.MethodPost, connectURL, nil, nil); err != nil {
 				return fmt.Errorf("failed to connect IPFS node %s to %s: %w", member.ID, other.ID, err)
 			}
+			s.Log.Info(fmt.Sprintf("IPFS node %s peered with %s (%s)", member.ID, other.ID, peerIDs[other.ID]))
 		}
 	}
+
+	s.Log.Info("IPFS nodes peered successfully")
 	return nil
 }