// Named Tmux Manager - Multi-agent orchestration for Claude Code, Codex, and Gemini in tiled tmux panes. Visual dashboards, command palette, context rotation, robot mode API, work assignment, safety system. Go CLI.
| name | ntm |
| description | Named Tmux Manager - Multi-agent orchestration for Claude Code, Codex, and Gemini in tiled tmux panes. Visual dashboards, command palette, context rotation, robot mode API, work assignment, safety system. Go CLI. |
A Go CLI that transforms tmux into a multi-agent command center for orchestrating Claude Code, Codex, and Gemini agents in parallel. Spawn, manage, and coordinate AI agents across tiled panes with stunning TUI, automated context rotation, and deep integrations with the Agent Flywheel ecosystem.
Managing multiple AI coding agents is painful:
NTM solves all of this with one session containing many agents, persistent across SSH disconnections.
# Install
curl -fsSL https://raw.githubusercontent.com/Dicklesworthstone/ntm/main/install.sh | bash
# Add shell integration
echo 'eval "$(ntm init zsh)"' >> ~/.zshrc && source ~/.zshrc
# Interactive tutorial
ntm tutorial
# Check dependencies
ntm deps -v
# Create multi-agent session
ntm spawn myproject --cc=2 --cod=1 --gmi=1
# Send prompt to all Claude agents
ntm send myproject --cc "Explore this codebase and summarize its architecture."
# Open command palette (or press F6 after `ntm bind`)
ntm palette myproject
ntm spawn myproject --cc=3 --cod=2 --gmi=1 # 3 Claude + 2 Codex + 1 Gemini
ntm quick myproject --template=go # Full project scaffold + agents
ntm create myproject --panes=10 # Empty panes only
ntm spawn myproject --profiles=architect,implementer,tester
| Flag | Agent | CLI Command |
|---|---|---|
--cc=N | Claude Code | claude |
--cod=N | Codex CLI | codex |
--gmi=N | Gemini CLI | gemini |
ntm add myproject --cc=2 # Add 2 more Claude agents
ntm add myproject --cod=1 --gmi=1 # Add mixed agents
ntm send myproject --cc "Implement user auth" # To all Claude
ntm send myproject --cod "Write unit tests" # To all Codex
ntm send myproject --gmi "Review and document" # To all Gemini
ntm send myproject --all "Review current state" # To ALL agents
ntm interrupt myproject # Ctrl+C to all
| Command | Alias | Description |
|---|---|---|
ntm list | lnt | List all tmux sessions |
ntm attach | rnt | Attach to session |
ntm status | snt | Show pane details with agent counts |
ntm view | vnt | Unzoom, tile layout, attach |
ntm zoom | znt | Zoom to specific pane |
ntm dashboard | dash, d | Interactive visual dashboard |
ntm kill | knt | Kill session (-f to force) |
Fuzzy-searchable TUI with pre-configured prompts:
ntm palette myproject # Open palette
ntm bind # Set up F6 keybinding
ntm bind --key=F5 # Use different key
Ctrl+P / Ctrl+F)| Key | Action |
|---|---|
↑/↓ or j/k | Navigate |
1-9 | Quick select |
Enter | Select command |
Esc | Back / Quit |
? | Help overlay |
Ctrl+P | Pin/unpin |
Ctrl+F | Favorite |
ntm dashboard myproject # Or: ntm dash myproject
r| Key | Action |
|---|---|
↑/↓ or j/k | Navigate panes |
1-9 | Quick select |
z or Enter | Zoom to pane |
r | Refresh |
c | View context |
m | Open Agent Mail |
q | Quit |
ntm copy myproject:1 # Copy specific pane
ntm copy myproject --all # Copy all panes
ntm copy myproject --cc # Copy Claude panes only
ntm copy myproject --pattern 'ERROR' # Filter by regex
ntm copy myproject --code # Extract code blocks only
ntm copy myproject --output out.txt # Save to file
ntm save myproject -o ~/logs # Save all outputs
ntm activity myproject --watch # Real-time activity
ntm health myproject # Health status
ntm watch myproject --cc # Stream output
ntm extract myproject --lang=go # Extract code blocks
ntm diff myproject cc_1 cod_1 # Compare panes
ntm grep 'error' myproject -C 3 # Search with context
ntm analytics --days 7 # Session statistics
ntm locks myproject --all-agents # File reservations
| State | Icon | Description |
|---|---|---|
| WAITING | ● | Idle, ready for work |
| GENERATING | ▶ | Producing output |
| THINKING | ◐ | Processing (no output yet) |
| ERROR | ✗ | Encountered error |
| STALLED | ◯ | Stopped unexpectedly |
ntm checkpoint save myproject -m "Before refactor"
ntm checkpoint list myproject
ntm checkpoint show myproject 20251210-143052
ntm checkpoint delete myproject 20251210-143052 -f
NTM monitors context usage and auto-rotates agents before exhausting context.
/compact or summarization first| Color | Usage | Status |
|---|---|---|
| Green | < 40% | Plenty of room |
| Yellow | 40-60% | Comfortable |
| Orange | 60-80% | Approaching threshold |
| Red | > 80% | Needs attention |
When context is compacted, NTM sends a recovery prompt:
[context_rotation.recovery]
enabled = true
prompt = "Reread AGENTS.md so it's still fresh in your mind. Use ultrathink."
include_bead_context = true # Include project state from bv
Machine-readable JSON output for AI agents and automation.
ntm --robot-status # Sessions, panes, agent states
ntm --robot-context=SESSION # Context window usage
ntm --robot-snapshot # Unified state: sessions + beads + mail
ntm --robot-tail=SESSION # Recent pane output
ntm --robot-inspect-pane=SESS # Detailed pane inspection
ntm --robot-files=SESSION # File changes with attribution
ntm --robot-metrics=SESSION # Session metrics
ntm --robot-plan # bv execution plan
ntm --robot-dashboard # Dashboard summary
ntm --robot-health # Project health
ntm --robot-send=SESSION --msg="Fix auth" --type=claude
ntm --robot-spawn=SESSION --spawn-cc=2 --spawn-wait
ntm --robot-interrupt=SESSION
ntm --robot-assign=SESSION --assign-beads=br-1,br-2
ntm --robot-replay=SESSION --replay-id=ID
ntm --robot-bead-claim=BEAD_ID --bead-assignee=agent
ntm --robot-bead-create --bead-title="Fix bug" --bead-type=bug
ntm --robot-bead-show=BEAD_ID
ntm --robot-bead-close=BEAD_ID --bead-close-reason="Fixed"
ntm --robot-cass-search="auth error" --cass-since=7d
ntm --robot-cass-context="how to implement auth"
ntm --robot-cass-status
| Code | Meaning |
|---|---|
0 | Success |
1 | Error |
2 | Unavailable/Not implemented |
Integration with BV for intelligent work assignment:
ntm work triage # Full triage with recommendations
ntm work triage --by-label # Group by domain
ntm work triage --quick # Quick wins only
ntm work alerts # Stale issues, priority drift, cycles
ntm work search "JWT auth" # Semantic search
ntm work impact src/api/*.go # Impact analysis
ntm work next # Single best next action
ntm --robot-assign=myproject --assign-strategy=balanced # Default
ntm --robot-assign=myproject --assign-strategy=speed # Maximize throughput
ntm --robot-assign=myproject --assign-strategy=quality # Best agent-task match
ntm --robot-assign=myproject --assign-strategy=dependency # Unblock downstream
| Agent | Best At |
|---|---|
| Claude | Analysis, refactoring, documentation, architecture |
| Codex | Feature implementation, bug fixes, quick tasks |
| Gemini | Documentation, analysis, features |
ntm profiles list # List profiles
ntm profiles show architect # Show details
ntm spawn myproject --profiles=architect,implementer,tester
ntm spawn myproject --profile-set=backend-team
architect, implementer, reviewer, tester, documenter
ntm mail send myproject --to GreenCastle "Review API changes"
ntm mail send myproject --all "Checkpoint: sync status"
ntm mail inbox myproject
ntm mail read myproject --agent BlueLake
ntm mail ack myproject 42
ntm hooks guard install # Prevent conflicting commits
ntm hooks guard uninstall
Multi-channel notifications for events:
[notifications]
enabled = true
events = ["agent.error", "agent.crashed", "agent.rate_limit"]
[notifications.desktop]
enabled = true
[notifications.webhook]
enabled = true
url = "https://hooks.slack.com/..."
agent.error, agent.crashed, agent.rate_limit, rotation.needed, session.created, session.killed, health.degraded
| Type | Severity | Description |
|---|---|---|
unhealthy | High | Agent enters unhealthy state |
degraded | Medium | Agent performance degrades |
rate_limited | Medium | API rate limit detected |
restart_failed | High | Restart attempt failed |
max_restarts | Critical | Restart limit exceeded |
ntm --robot-alerts
ntm --robot-dismiss-alert=ALERT_ID
# ~/.config/ntm/hooks.toml
[[command_hooks]]
event = "post-spawn"
command = "notify-send 'NTM' 'Agents spawned'"
[[command_hooks]]
event = "pre-send"
command = "echo \"$(date): $NTM_MESSAGE\" >> ~/.ntm-send.log"
pre-spawn, post-spawn, pre-send, post-send, pre-add, post-add, pre-shutdown, post-shutdown
Blocks dangerous commands from AI agents:
ntm safety status # Protection status
ntm safety check "git reset --hard"
ntm safety install # Install git wrapper + Claude hook
ntm safety uninstall
| Pattern | Risk | Action |
|---|---|---|
git reset --hard | Loses uncommitted changes | Block |
git push --force | Overwrites remote history | Block |
rm -rf / | Catastrophic deletion | Block |
DROP TABLE | Database destruction | Block |
ntm send myproject --cc "design the database schema"
ntm send myproject --cod "implement the models"
ntm send myproject --gmi "write tests"
ntm send myproject --all "implement a rate limiter"
ntm view myproject # Compare side-by-side
ntm send myproject --cc "implement feature X"
ntm send myproject --cod "review Claude's code"
ntm send myproject --gmi "write tests for edge cases"
ntm config init # Create ~/.config/ntm/config.toml
ntm config show # Show current config
ntm config project init # Create .ntm/config.toml in project
projects_base = "~/Developer"
[agents]
claude = 'claude --dangerously-skip-permissions'
codex = "codex --dangerously-bypass-approvals-and-sandbox"
gemini = "gemini --yolo"
[tmux]
default_panes = 10
palette_key = "F6"
[context_rotation]
enabled = true
warning_threshold = 0.80
rotate_threshold = 0.95
| Variable | Description |
|---|---|
NTM_PROJECTS_BASE | Base directory for projects |
NTM_THEME | Color theme: auto, mocha, latte, nord, plain |
NTM_ICONS | Icon set: nerd, unicode, ascii |
NTM_REDUCE_MOTION | Disable animations |
NTM_PROFILE | Enable performance profiling |
| Theme | Description |
|---|---|
auto | Detect light/dark |
mocha | Default dark, warm |
latte | Light variant |
nord | Arctic-inspired |
plain | No color |
| Agent | Color |
|---|---|
| Claude | Mauve (Purple) |
| Codex | Blue |
| Gemini | Yellow |
| User | Green |
| Width | Behavior |
|---|---|
| <120 cols | Stacked layout |
| 120-199 cols | List/detail split |
| 200-239 cols | Wider gutters |
| 240+ cols | Full detail |
Pattern: <project>__<agent>_<number>
myproject__cc_1 - First Claudemyproject__cod_2 - Second Codexmyproject__gmi_1 - First GeminiStatus indicators: C = Claude, X = Codex, G = Gemini, U = User
After eval "$(ntm init zsh)":
| Category | Aliases |
|---|---|
| Agent Launch | cc, cod, gmi |
| Session | cnt, sat, qps |
| Agent Mgmt | ant, bp, int |
| Navigation | rnt, lnt, snt, vnt, znt |
| Dashboard | dash, d |
| Output | cpnt, svnt |
| Utilities | ncp, knt, cad |
# One-liner (recommended)
curl -fsSL https://raw.githubusercontent.com/Dicklesworthstone/ntm/main/install.sh | bash
# Homebrew
brew install dicklesworthstone/tap/ntm
# Go install
go install github.com/Dicklesworthstone/ntm/cmd/ntm@latest
# Docker
docker pull ghcr.io/dicklesworthstone/ntm:latest
ntm upgrade # Check and install updates
ntm upgrade --check # Check only
ntm upgrade --yes # Auto-confirm
| Keys | Action |
|---|---|
Ctrl+B, D | Detach |
Ctrl+B, [ | Scroll/copy mode |
Ctrl+B, z | Toggle zoom |
Ctrl+B, Arrow | Navigate panes |
F6 | Open NTM palette (after ntm bind) |
| Tool | Integration |
|---|---|
| Agent Mail | Message routing, file reservations, pre-commit guard |
| BV | Work distribution, triage, assignment strategies |
| CASS | Search past sessions via robot mode |
| CM | Procedural memory for agent handoffs |
| DCG | Safety system integration |
| UBS | Auto-scanning on file changes |
MCP Agent Mail - Mail-like coordination layer for multi-agent workflows. Identities, inbox/outbox, file reservations, contact policies, threaded messaging, pre-commit guard, Human Overseer, static exports, disaster recovery. Git+SQLite backed. Python/FastMCP.
Jeffrey Emanuel's multi-agent implementation workflow using NTM, Agent Mail, Beads, and BV. The execution phase that follows planning and bead creation. Includes exact prompts used.
Converting markdown plans into beads (tasks with dependencies) and polishing them until they're implementation-ready. The bridge between planning and agent swarm execution. Includes exact prompts used.
Beads Viewer - Graph-aware triage engine for Beads projects. Computes PageRank, betweenness, critical path, and cycles. Use --robot-* flags for AI agents.
Remove telltale signs of AI-generated 'slop' writing from README files and documentation. Make your docs sound authentically human.
Find opportunities to improve web application code using TanStack libraries (Query, Table, Form, Router, etc.). Avoid man-with-hammer syndrome by applying TanStack after vanilla implementation works.