Zorai is a persistent, multi-agent, auditable, learning execution platform where the daemon owns work, memory, approvals, tools, and long-running goals.
Zorai is a persistent, multi-agent, auditable, learning execution platform where the daemon owns work, memory, approvals, tools, and long-running goals.
Read our docs
Fully agentic daemon runtime for durable AI work.
Zorai is a persistent, multi-agent, auditable, learning execution platform where the daemon owns work, memory, approvals, tools, and long-running goals.
Official website: https://zorai.app
Zorai keeps the agent runtime, operator surfaces, tools, memory, and governed execution in one place. Threads, workspace tasks, approvals, and goal runs live in the daemon, so work can keep moving even when the UI closes.
The name comes from Zora: mythic but clean, suggesting dawn, awakening, watchfulness, and beginnings. It nods to Slavic dawn and light motifs without leaning on a brittle one-to-one mythological claim. The final i completes the AI identity.
In practice that means:
Before you jump into installation and setup, read the operational guidance in docs/best-practices.md. It covers model role selection, compaction strategy, governance usage, cost control, and the day-to-day habits that make zorai work well.
Linux/macOS:
curl -fsSL https://raw.githubusercontent.com/mkurman/zorai/main/scripts/install.sh | sh
zorai --helpWindows PowerShell, run as Administrator:
irm https://raw.githubusercontent.com/mkurman/zorai/main/scripts/install.ps1 | iex
zorai --help
zorai guiThe quick installer downloads the same native release bundle family that the npm package uses, installs the binaries into ~/.local/bin on Linux/macOS or C:\Program Files\zorai on Windows by default, and works without Node.js or npm. Windows x64 and Windows ARM64 use architecture-matched bundles, and every bundle includes the full app surface: CLI, daemon, TUI, MCP/gateway binaries, and the desktop GUI launcher (zorai-desktop.exe on Windows).
After a quick-install or direct-binary install, zorai upgrade reuses the direct installer path for the current install directory. If you installed through npm, zorai upgrade continues to use npm.
npm install -g zor-ai
zorai --help
# or install locally in a project:
npm install zor-ai
npx zor-ai --helpIf npm install -g zor-ai succeeds but zorai is still not found on macOS, your npm global bin directory is not on PATH yet:
export PATH="$(npm config get prefix)/bin:$PATH"
exec $SHELL -l
zorai --helpIf the bin directory is already on PATH, opening a new shell is still useful because zsh and bash can cache command lookups.
When zorai is installed through npm, zorai upgrade upgrades through npm as well.
The npm package also installs the full platform bundle on every supported platform. On Windows, npm install -g zor-ai installs zorai-desktop.exe alongside the CLI binaries so zorai gui works after install.
git clone https://github.com/mkurman/zorai.git
cd zorai
# 1. Start the daemon
cargo run --release --bin zorai-daemon
# 2. Launch the TUI in another terminal
cargo run --release --bin zorai setup
# or
cargo run --release --bin zorai-tui
# 3. Or launch the desktop app
cd frontend && npm install && npm run dev:electronIf you want to test the core loop fast, start a goal run and give Swarog a concrete objective: investigate a failing build, prepare a release checklist, or trace a bug across a workspace.
zorai is for operators who want agents that stay with durable work and control surfaces that make long-running execution visible instead of mysterious.
Terminals remain first-class tools, especially for engineering work, but they are no longer the product's core premise. The center is the agentic runtime: planning, governed tool use, memory, collaboration, review, and durable goal execution.
The TUI is a keyboard-first control room for the daemon with mouse support.
Thread command reference: docs/tui/agent-directives.md
zorai gives its daemon-side agents a slightly mythic face, but the work stays concrete.
Use @agent ... to add or update a visible thread participant on the current thread.
Examples:
@weles verify claims before answering@swarozyc review svarog's implementation and step in when needed@perun drive the risky migration plan to a concrete decision@mokosh stabilize this workspace and clean up the rough edges@dazhbog turn this discussion into a clear plan and operator updateBehavior:
@agent ... updates that participant instead of creating duplicatesUse !agent ... for hidden internal DM.
Examples:
!weles check whether this is risky!radogost compare these two approaches before we answer!perun assess the operational blast radius before we proceed!dazhbog summarize the strongest next move in plain languageBehavior:
Use handoff when another agent should own future replies in the thread.
Behavior:
Builtin personas such as @swarozyc, @radogost, @domowoj, @swietowit, @perun, @mokosh, and @dazhbog need their own provider and model configuration before they can participate or receive internal DM.
If one of those personas is available but not configured yet:
@agent ... or !agent ... request automatically@veles is accepted as an alias for @weles, matching the existing swarog and svarog alias handling.
If the target is not a known built-in persona or registered sub-agent alias, the leading @agent form is not treated as an agent directive and remains a normal message/file-mention path.
Together they give the system a little presence without hiding what it is doing.
Zorai supports voice workflows in both TUI and desktop app.
Ctrl+L (record/transcribe), Ctrl+P (speak selected/latest assistant message), Ctrl+S (stop playback)extra.audio_*Zorai can export daemon-wide agent turns, model phases, and tool calls to MLflow 3.6+ over OTLP/HTTP. Tracing is disabled by default, privacy-controlled, bounded, and fail-open. See docs/mlflow-tracing.md for local server setup, Desktop/TUI configuration, capture modes, scopes, encrypted headers, and troubleshooting.
No open issues yet, or sync has not completed.