Installation

Quick install

Native binaries without npm.

curl -fsSL https://raw.githubusercontent.com/mkurman/tamux/main/scripts/install.sh | sh
tamux --help

Source install

Full daemon + Electron + TUI workflow.

git clone https://github.com/mkurman/tamux.git
cd tamux
./scripts/setup.sh --check --profile source

Requires Rust, Node.js, npm, git, and uv. Start the daemon first; everything else reconnects to it.

First-run checklist

  1. Run ./scripts/setup.sh --check --profile source
  2. Start daemon: cargo run --release --bin tamux-daemon
  3. Open your preferred surface: Electron, TUI, CLI, or MCP
  4. Verify runtime points at the daemon backend
  5. Inspect goal, task, and approval surfaces
Key point: without the daemon, the system has no source of truth.

First goal

Run a bounded, concrete objective to see planning, child tasks, approvals, and reflection in action.

Investigate why the nightly Rust build is failing, identify the root cause,
propose the smallest safe fix, and capture any reusable workflow as a skill.

What to expect

  • Run enters queued, then planning
  • Daemon generates plan and child tasks
  • Risky operations pause for approval
  • Run ends with reflection, memory update, or skill

Choosing surfaces

SurfaceBest forTypical use
ElectronDesktop overviewGoals, settings, plugin UI, multi-pane
TUIKeyboard-first, SSHInspect sessions, goals, tasks, approvals
CLIScriptingAutomation, session management, health
MCPExternal integrationExpose tamux to outside tools
GatewayChat platformsDiscord, Slack, Telegram, WhatsApp

All surfaces are clients of the same daemon state.

Daily habits

Durable threads

One thread per objective. Continue when the mission is the same. New thread only when the objective changes.

Goal runners for real work

Multi-step, long-running, or approval-heavy tasks benefit from goals over chat threads.

Explicit success criteria

Define done, constraints, and verification. tamux rewards specificity.

Daemon owns state

UI closing is not failure. Work continues and reconnects.

Model roles

  • Svarog: primary execution, synthesis, goal runs
  • Weles: review, compaction, governance checks
  • Rarog: lightweight operator assistance
Don't starve governance: Weles needs enough capability for reliable review.

Scenarios

Bug hunt

Goal runner for root-cause analysis, smallest fix, local verification.

Release prep

Thread for discussion, goal for checklist execution and packaging.

Research task

Goals for multi-search, comparison, artifacts, branches.

Remote / SSH

TUI as control room, daemon running remotely, reconnect as needed.

When not to use

Use simple chat for short, disposable tasks without queueing, goals, approvals, or durable artifacts. tamux shines when continuity matters.