diff --git a/internal/cmd/up.go b/internal/cmd/up.go
index e4820fbf..e28245ea 100644
--- a/internal/cmd/up.go
+++ b/internal/cmd/up.go
@@ -6,6 +6,7 @@ import (
 	"os/exec"
 	"path/filepath"
 	"strings"
+	"sync"
 	"time"
 
 	"github.com/spf13/cobra"
@@ -18,12 +19,23 @@ import (
 	"github.com/steveyegge/gastown/internal/mayor"
 	"github.com/steveyegge/gastown/internal/polecat"
 	"github.com/steveyegge/gastown/internal/refinery"
+	"github.com/steveyegge/gastown/internal/rig"
 	"github.com/steveyegge/gastown/internal/style"
 	"github.com/steveyegge/gastown/internal/tmux"
 	"github.com/steveyegge/gastown/internal/witness"
 	"github.com/steveyegge/gastown/internal/workspace"
 )
 
+// agentStartResult holds the result of starting an agent.
+type agentStartResult struct {
+	name   string // Display name like "Witness (gastown)"
+	ok     bool   // Whether start succeeded
+	detail string // Status detail (session name or error)
+}
+
+// maxConcurrentAgentStarts limits parallel agent startups to avoid resource exhaustion.
+const maxConcurrentAgentStarts = 10
+
 var upCmd = &cobra.Command{
 	Use:     "up",
 	GroupID: GroupServices,
@@ -70,89 +82,108 @@ func runUp(cmd *cobra.Command, args []string) error {
 
 	allOK := true
 
-	// 1. Daemon (Go process)
-	if err := ensureDaemon(townRoot); err != nil {
-		printStatus("Daemon", false, err.Error())
-		allOK = false
-	} else {
-		running, pid, _ := daemon.IsRunning(townRoot)
-		if running {
-			printStatus("Daemon", true, fmt.Sprintf("PID %d", pid))
-		}
-	}
-
-	// 2. Deacon (Claude agent)
-	deaconMgr := deacon.NewManager(townRoot)
-	if err := deaconMgr.Start(""); err != nil {
-		if err == deacon.ErrAlreadyRunning {
-			printStatus("Deacon", true, deaconMgr.SessionName())
-		} else {
-			printStatus("Deacon", false, err.Error())
-			allOK = false
-		}
-	} else {
-		printStatus("Deacon", true, deaconMgr.SessionName())
-	}
-
-	// 3. Mayor (Claude agent)
-	mayorMgr := mayor.NewManager(townRoot)
-	if err := mayorMgr.Start(""); err != nil {
-		if err == mayor.ErrAlreadyRunning {
-			printStatus("Mayor", true, mayorMgr.SessionName())
-		} else {
-			printStatus("Mayor", false, err.Error())
-			allOK = false
-		}
-	} else {
-		printStatus("Mayor", true, mayorMgr.SessionName())
-	}
-
-	// 4. Witnesses (one per rig)
+	// Discover rigs early so we can prefetch while daemon/deacon/mayor start
 	rigs := discoverRigs(townRoot)
-	for _, rigName := range rigs {
-		_, r, err := getRig(rigName)
-		if err != nil {
-			printStatus(fmt.Sprintf("Witness (%s)", rigName), false, err.Error())
-			allOK = false
-			continue
-		}
 
-		mgr := witness.NewManager(r)
-		if err := mgr.Start(false, "", nil); err != nil {
-			if err == witness.ErrAlreadyRunning {
-				printStatus(fmt.Sprintf("Witness (%s)", rigName), true, mgr.SessionName())
+	// Start daemon, deacon, mayor, and rig prefetch in parallel
+	var daemonErr error
+	var daemonPID int
+	var deaconResult, mayorResult agentStartResult
+	var prefetchedRigs map[string]*rig.Rig
+	var rigErrors map[string]error
+
+	var startupWg sync.WaitGroup
+	startupWg.Add(4)
+
+	// 1. Daemon (Go process)
+	go func() {
+		defer startupWg.Done()
+		if err := ensureDaemon(townRoot); err != nil {
+			daemonErr = err
+		} else {
+			running, pid, _ := daemon.IsRunning(townRoot)
+			if running {
+				daemonPID = pid
+			}
+		}
+	}()
+
+	// 2. Deacon
+	go func() {
+		defer startupWg.Done()
+		deaconMgr := deacon.NewManager(townRoot)
+		if err := deaconMgr.Start(""); err != nil {
+			if err == deacon.ErrAlreadyRunning {
+				deaconResult = agentStartResult{name: "Deacon", ok: true, detail: deaconMgr.SessionName()}
 			} else {
-				printStatus(fmt.Sprintf("Witness (%s)", rigName), false, err.Error())
-				allOK = false
+				deaconResult = agentStartResult{name: "Deacon", ok: false, detail: err.Error()}
 			}
 		} else {
-			printStatus(fmt.Sprintf("Witness (%s)", rigName), true, mgr.SessionName())
+			deaconResult = agentStartResult{name: "Deacon", ok: true, detail: deaconMgr.SessionName()}
+		}
+	}()
+
+	// 3. Mayor
+	go func() {
+		defer startupWg.Done()
+		mayorMgr := mayor.NewManager(townRoot)
+		if err := mayorMgr.Start(""); err != nil {
+			if err == mayor.ErrAlreadyRunning {
+				mayorResult = agentStartResult{name: "Mayor", ok: true, detail: mayorMgr.SessionName()}
+			} else {
+				mayorResult = agentStartResult{name: "Mayor", ok: false, detail: err.Error()}
+			}
+		} else {
+			mayorResult = agentStartResult{name: "Mayor", ok: true, detail: mayorMgr.SessionName()}
+		}
+	}()
+
+	// 4. Prefetch rig configs (overlaps with daemon/deacon/mayor startup)
+	go func() {
+		defer startupWg.Done()
+		prefetchedRigs, rigErrors = prefetchRigs(rigs)
+	}()
+
+	startupWg.Wait()
+
+	// Print daemon/deacon/mayor results
+	if daemonErr != nil {
+		printStatus("Daemon", false, daemonErr.Error())
+		allOK = false
+	} else if daemonPID > 0 {
+		printStatus("Daemon", true, fmt.Sprintf("PID %d", daemonPID))
+	}
+	printStatus(deaconResult.name, deaconResult.ok, deaconResult.detail)
+	if !deaconResult.ok {
+		allOK = false
+	}
+	printStatus(mayorResult.name, mayorResult.ok, mayorResult.detail)
+	if !mayorResult.ok {
+		allOK = false
+	}
+
+	// 5 & 6. Witnesses and Refineries (using prefetched rigs)
+	witnessResults, refineryResults := startRigAgentsWithPrefetch(rigs, prefetchedRigs, rigErrors)
+
+	// Print results in order: all witnesses first, then all refineries
+	for _, rigName := range rigs {
+		if result, ok := witnessResults[rigName]; ok {
+			printStatus(result.name, result.ok, result.detail)
+			if !result.ok {
+				allOK = false
+			}
+		}
+	}
+	for _, rigName := range rigs {
+		if result, ok := refineryResults[rigName]; ok {
+			printStatus(result.name, result.ok, result.detail)
+			if !result.ok {
+				allOK = false
+			}
 		}
 	}
 
-	// 5. Refineries (one per rig)
-	for _, rigName := range rigs {
-		_, r, err := getRig(rigName)
-		if err != nil {
-			printStatus(fmt.Sprintf("Refinery (%s)", rigName), false, err.Error())
-			allOK = false
-			continue
-		}
-
-		mgr := refinery.NewManager(r)
-		if err := mgr.Start(false, ""); err != nil {
-			if err == refinery.ErrAlreadyRunning {
-				printStatus(fmt.Sprintf("Refinery (%s)", rigName), true, mgr.SessionName())
-			} else {
-				printStatus(fmt.Sprintf("Refinery (%s)", rigName), false, err.Error())
-				allOK = false
-			}
-		} else {
-			printStatus(fmt.Sprintf("Refinery (%s)", rigName), true, mgr.SessionName())
-		}
-	}
-
-	// 6. Crew (if --restore)
+	// 7. Crew (if --restore)
 	if upRestore {
 		for _, rigName := range rigs {
 			crewStarted, crewErrors := startCrewFromSettings(townRoot, rigName)
@@ -249,6 +280,182 @@ func ensureDaemon(townRoot string) error {
 	return nil
 }
 
+// rigPrefetchResult holds the result of loading a single rig config.
+type rigPrefetchResult struct {
+	index int
+	rig   *rig.Rig
+	err   error
+}
+
+// prefetchRigs loads all rig configs in parallel for faster agent startup.
+// Returns a map of rig name to loaded Rig, and any errors encountered.
+func prefetchRigs(rigNames []string) (map[string]*rig.Rig, map[string]error) {
+	n := len(rigNames)
+	if n == 0 {
+		return make(map[string]*rig.Rig), make(map[string]error)
+	}
+
+	// Use channel to collect results without locking
+	results := make(chan rigPrefetchResult, n)
+
+	for i, name := range rigNames {
+		go func(idx int, rigName string) {
+			_, r, err := getRig(rigName)
+			results <- rigPrefetchResult{index: idx, rig: r, err: err}
+		}(i, name)
+	}
+
+	// Collect results - pre-allocate maps with capacity
+	rigs := make(map[string]*rig.Rig, n)
+	errors := make(map[string]error)
+
+	for i := 0; i < n; i++ {
+		res := <-results
+		name := rigNames[res.index]
+		if res.err != nil {
+			errors[name] = res.err
+		} else {
+			rigs[name] = res.rig
+		}
+	}
+
+	return rigs, errors
+}
+
+// agentTask represents a unit of work for the agent worker pool.
+type agentTask struct {
+	rigName   string
+	rigObj    *rig.Rig
+	isWitness bool // true for witness, false for refinery
+}
+
+// agentResultMsg carries result back from worker to collector.
+type agentResultMsg struct {
+	rigName   string
+	isWitness bool
+	result    agentStartResult
+}
+
+// startRigAgentsParallel starts all Witnesses and Refineries concurrently.
+// Discovers and prefetches rigs internally. For use when rigs aren't pre-loaded.
+func startRigAgentsParallel(rigNames []string) (witnessResults, refineryResults map[string]agentStartResult) {
+	prefetchedRigs, rigErrors := prefetchRigs(rigNames)
+	return startRigAgentsWithPrefetch(rigNames, prefetchedRigs, rigErrors)
+}
+
+// startRigAgentsWithPrefetch starts all Witnesses and Refineries using pre-loaded rig configs.
+// Uses a worker pool with fixed goroutine count to limit concurrency and reduce overhead.
+func startRigAgentsWithPrefetch(rigNames []string, prefetchedRigs map[string]*rig.Rig, rigErrors map[string]error) (witnessResults, refineryResults map[string]agentStartResult) {
+	n := len(rigNames)
+	witnessResults = make(map[string]agentStartResult, n)
+	refineryResults = make(map[string]agentStartResult, n)
+
+	if n == 0 {
+		return
+	}
+
+	// Record errors for rigs that failed to load
+	for rigName, err := range rigErrors {
+		errDetail := err.Error()
+		witnessResults[rigName] = agentStartResult{
+			name:   "Witness (" + rigName + ")",
+			ok:     false,
+			detail: errDetail,
+		}
+		refineryResults[rigName] = agentStartResult{
+			name:   "Refinery (" + rigName + ")",
+			ok:     false,
+			detail: errDetail,
+		}
+	}
+
+	numTasks := len(prefetchedRigs) * 2 // witness + refinery per rig
+	if numTasks == 0 {
+		return
+	}
+
+	// Task channel and result channel
+	tasks := make(chan agentTask, numTasks)
+	results := make(chan agentResultMsg, numTasks)
+
+	// Start fixed worker pool (bounded by maxConcurrentAgentStarts)
+	numWorkers := maxConcurrentAgentStarts
+	if numTasks < numWorkers {
+		numWorkers = numTasks
+	}
+
+	var wg sync.WaitGroup
+	for i := 0; i < numWorkers; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			for task := range tasks {
+				var result agentStartResult
+				if task.isWitness {
+					result = upStartWitness(task.rigName, task.rigObj)
+				} else {
+					result = upStartRefinery(task.rigName, task.rigObj)
+				}
+				results <- agentResultMsg{
+					rigName:   task.rigName,
+					isWitness: task.isWitness,
+					result:    result,
+				}
+			}
+		}()
+	}
+
+	// Enqueue all tasks
+	for rigName, r := range prefetchedRigs {
+		tasks <- agentTask{rigName: rigName, rigObj: r, isWitness: true}
+		tasks <- agentTask{rigName: rigName, rigObj: r, isWitness: false}
+	}
+	close(tasks)
+
+	// Close results channel when workers are done
+	go func() {
+		wg.Wait()
+		close(results)
+	}()
+
+	// Collect results - no locking needed, single goroutine collects
+	for msg := range results {
+		if msg.isWitness {
+			witnessResults[msg.rigName] = msg.result
+		} else {
+			refineryResults[msg.rigName] = msg.result
+		}
+	}
+
+	return
+}
+
+// upStartWitness starts a witness for the given rig and returns a result struct.
+func upStartWitness(rigName string, r *rig.Rig) agentStartResult {
+	name := "Witness (" + rigName + ")"
+	mgr := witness.NewManager(r)
+	if err := mgr.Start(false, "", nil); err != nil {
+		if err == witness.ErrAlreadyRunning {
+			return agentStartResult{name: name, ok: true, detail: mgr.SessionName()}
+		}
+		return agentStartResult{name: name, ok: false, detail: err.Error()}
+	}
+	return agentStartResult{name: name, ok: true, detail: mgr.SessionName()}
+}
+
+// upStartRefinery starts a refinery for the given rig and returns a result struct.
+func upStartRefinery(rigName string, r *rig.Rig) agentStartResult {
+	name := "Refinery (" + rigName + ")"
+	mgr := refinery.NewManager(r)
+	if err := mgr.Start(false, ""); err != nil {
+		if err == refinery.ErrAlreadyRunning {
+			return agentStartResult{name: name, ok: true, detail: mgr.SessionName()}
+		}
+		return agentStartResult{name: name, ok: false, detail: err.Error()}
+	}
+	return agentStartResult{name: name, ok: true, detail: mgr.SessionName()}
+}
+
 // discoverRigs finds all rigs in the town.
 func discoverRigs(townRoot string) []string {
 	var rigs []string
diff --git a/internal/cmd/up_test.go b/internal/cmd/up_test.go
new file mode 100644
index 00000000..5d8ef51b
--- /dev/null
+++ b/internal/cmd/up_test.go
@@ -0,0 +1,216 @@
+package cmd
+
+import (
+	"fmt"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/steveyegge/gastown/internal/rig"
+)
+
+func TestAgentStartResult_Fields(t *testing.T) {
+	result := agentStartResult{
+		name:   "Witness (gastown)",
+		ok:     true,
+		detail: "gt-gastown-witness",
+	}
+
+	if result.name != "Witness (gastown)" {
+		t.Errorf("name = %q, want %q", result.name, "Witness (gastown)")
+	}
+	if !result.ok {
+		t.Error("ok should be true")
+	}
+	if result.detail != "gt-gastown-witness" {
+		t.Errorf("detail = %q, want %q", result.detail, "gt-gastown-witness")
+	}
+}
+
+func TestMaxConcurrentAgentStarts_Constant(t *testing.T) {
+	// Verify the constant is set to a reasonable value
+	if maxConcurrentAgentStarts < 1 {
+		t.Errorf("maxConcurrentAgentStarts = %d, should be >= 1", maxConcurrentAgentStarts)
+	}
+	if maxConcurrentAgentStarts > 100 {
+		t.Errorf("maxConcurrentAgentStarts = %d, should be <= 100 to prevent resource exhaustion", maxConcurrentAgentStarts)
+	}
+}
+
+func TestSemaphoreLimitsConcurrency(t *testing.T) {
+	// Test that a semaphore pattern properly limits concurrency
+	const maxConcurrent = 3
+	const totalTasks = 10
+
+	sem := make(chan struct{}, maxConcurrent)
+	var wg sync.WaitGroup
+	var maxObserved int32
+	var current int32
+
+	for i := 0; i < totalTasks; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+
+			// Acquire semaphore
+			sem <- struct{}{}
+			defer func() { <-sem }()
+
+			// Track concurrent count
+			cur := atomic.AddInt32(&current, 1)
+			defer atomic.AddInt32(&current, -1)
+
+			// Update max observed
+			for {
+				max := atomic.LoadInt32(&maxObserved)
+				if cur <= max || atomic.CompareAndSwapInt32(&maxObserved, max, cur) {
+					break
+				}
+			}
+
+			// Simulate work
+			time.Sleep(10 * time.Millisecond)
+		}()
+	}
+
+	wg.Wait()
+
+	if maxObserved > maxConcurrent {
+		t.Errorf("max concurrent = %d, should not exceed %d", maxObserved, maxConcurrent)
+	}
+}
+
+func TestStartRigAgentsParallel_EmptyRigs(t *testing.T) {
+	// Test with empty rig list - should return empty maps without error
+	witnessResults, refineryResults := startRigAgentsParallel([]string{})
+
+	if len(witnessResults) != 0 {
+		t.Errorf("witnessResults should be empty, got %d entries", len(witnessResults))
+	}
+	if len(refineryResults) != 0 {
+		t.Errorf("refineryResults should be empty, got %d entries", len(refineryResults))
+	}
+}
+
+func TestStartRigAgentsWithPrefetch_EmptyRigs(t *testing.T) {
+	// Test with empty inputs
+	witnessResults, refineryResults := startRigAgentsWithPrefetch(
+		[]string{},
+		make(map[string]*rig.Rig),
+		make(map[string]error),
+	)
+
+	if len(witnessResults) != 0 {
+		t.Errorf("witnessResults should be empty, got %d entries", len(witnessResults))
+	}
+	if len(refineryResults) != 0 {
+		t.Errorf("refineryResults should be empty, got %d entries", len(refineryResults))
+	}
+}
+
+func TestStartRigAgentsWithPrefetch_RecordsErrors(t *testing.T) {
+	// Test that rig errors are properly recorded
+	rigErrors := map[string]error{
+		"badrig": fmt.Errorf("rig not found"),
+	}
+
+	witnessResults, refineryResults := startRigAgentsWithPrefetch(
+		[]string{"badrig"},
+		make(map[string]*rig.Rig),
+		rigErrors,
+	)
+
+	if len(witnessResults) != 1 {
+		t.Errorf("witnessResults should have 1 entry, got %d", len(witnessResults))
+	}
+	if result, ok := witnessResults["badrig"]; !ok {
+		t.Error("witnessResults should have badrig entry")
+	} else if result.ok {
+		t.Error("badrig witness result should not be ok")
+	}
+
+	if len(refineryResults) != 1 {
+		t.Errorf("refineryResults should have 1 entry, got %d", len(refineryResults))
+	}
+	if result, ok := refineryResults["badrig"]; !ok {
+		t.Error("refineryResults should have badrig entry")
+	} else if result.ok {
+		t.Error("badrig refinery result should not be ok")
+	}
+}
+
+func TestPrefetchRigs_Empty(t *testing.T) {
+	// Test with empty rig list
+	rigs, errors := prefetchRigs([]string{})
+
+	if len(rigs) != 0 {
+		t.Errorf("rigs should be empty, got %d entries", len(rigs))
+	}
+	if len(errors) != 0 {
+		t.Errorf("errors should be empty, got %d entries", len(errors))
+	}
+}
+
+func TestWorkerPoolLimitsConcurrency(t *testing.T) {
+	// Test that a worker pool pattern properly limits concurrency
+	const numWorkers = 3
+	const numTasks = 15
+
+	tasks := make(chan int, numTasks)
+	results := make(chan int, numTasks)
+
+	var maxObserved int32
+	var current int32
+
+	// Start worker pool
+	var wg sync.WaitGroup
+	for i := 0; i < numWorkers; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			for range tasks {
+				// Track concurrent count
+				cur := atomic.AddInt32(&current, 1)
+
+				// Update max observed
+				for {
+					max := atomic.LoadInt32(&maxObserved)
+					if cur <= max || atomic.CompareAndSwapInt32(&maxObserved, max, cur) {
+						break
+					}
+				}
+
+				// Simulate work
+				time.Sleep(5 * time.Millisecond)
+
+				atomic.AddInt32(&current, -1)
+				results <- 1
+			}
+		}()
+	}
+
+	// Enqueue tasks
+	for i := 0; i < numTasks; i++ {
+		tasks <- i
+	}
+	close(tasks)
+
+	// Wait for workers and collect results
+	go func() {
+		wg.Wait()
+		close(results)
+	}()
+
+	count := 0
+	for range results {
+		count++
+	}
+
+	if count != numTasks {
+		t.Errorf("expected %d results, got %d", numTasks, count)
+	}
+	if maxObserved > numWorkers {
+		t.Errorf("max concurrent = %d, should not exceed %d workers", maxObserved, numWorkers)
+	}
+}