Loading…
Free preview · for developers who run AI coding agents
Run a whole agent fleet in one workspace
nala is a local desktop app that puts Claude, Codex, Cursor, Grok, and the rest side by side with shared tasks and messages. Stop pasting context between terminal tabs. The preview costs nothing.
powershell -NoProfile -ExecutionPolicy Bypass -Command "irm 'https://nalarun.com/install.ps1' | iex"
Free preview · v3.24.27
Built for people already paying for coding agents
You keep Claude, Codex, Cursor, Grok, Kimi, and the rest. nala does not resell models. The app is free in preview; you only pay the providers you already use.
ClaudeCodexDevinCursorGrokKimi+ Custom CLI
vs five terminal tabs
Separate terminals make you the messenger.
nala makes the fleet talk.
Five agents in five windows means you copy context, restate the plan, and hope nobody stomps the same file. In nala they share a task board, send durable messages, and wait for your approval when something actually needs a human. That is the product, not a plugin bolted onto a chat panel.
Tabs
You relay every handoff by hand.
nala panes
Agents see shared tasks and state.
Your role
Approve the hard calls, not every paste.
FIRST HOUR
Install once. Connect the agents you already have.
Download the free preview, point nala at the coding tools on your machine, and open a workspace. Keys go from your laptop to the provider. Nothing is proxied through us.
Install · One package for desktop, CLI, and the resident agent.
Connect · Claude, Codex, Cursor, Grok, Kimi, or a custom CLI.
Conduct · Assign work, watch handoffs, approve only what matters.
powershell -NoProfile -ExecutionPolicy Bypass -Command "irm 'https://nalarun.com/install.ps1' | iex"Free preview · then type nala or nala desktop
01FLEXIBILITY
Extends your existing subscriptions
The supported coding tools and provider clients you already use remain the engine. Connect their local CLIs or a custom adapter; nala coordinates them without reselling inference or requiring a nala-hosted model backend.
CLAUDECODEXDEVINCURSORGROKKIMI+ CUSTOM
02DURABILITY
Terminals that survive reboots
Sessions live in a detached daemon: close the app or reboot the machine and the daemon can restore your terminals, scrollback, and agent conversations. Suspended sessions are kept for up to seven days.
DAEMONSCROLLBACKRECOVERYOSC 133
03INTEGRATION
An agent extensibility harness
nala registers itself as an MCP server with Claude, Codex, and Gemini: 30+ tools for workspace control, terminal routing, and browser automation, handed to every agent you run. Surgical registration that never clobbers your own config. Your agents arrive already knowing how to drive the machine.
MCPPLAYWRIGHTTOOLSPERMISSIONS
04WORKFLOW
Built-In Skills
nala ships with reusable workflows for planning, debugging, reviewing, debating, documenting, and breaking work into tickets. 40+ predefined agent roles across engineering, design, and PM.
PLANEXECUTEREVIEWEXPLAIN
THE A2A PROTOCOL
The fleet talks. You conduct.
Stop being the messenger between chat windows. Agents talk over a durable protocol (send, ask, reply, inbox), so a reviewer can question the builder, a lead can task its department, and you can put one agent in charge of others.
And no agent mints another agent without your approval: spawn requests are proposals until you accept them, and each approval is single-use.
Durable envelopesAcknowledged end-to-endIdempotent deliveryLoop preventionSingle-use spawn approvals
nala · live sessionFeature build · simulation
Off
Agents work alone. Messages queue, nothing moves. Your fleet, on mute.
Connect
Agents ask, reply, and hand off through the inbox. Envelopes survive restarts and are delivered with idempotent retry (at-least-once, not silent loss).
Orchestrate
Full choreography: reviews, delegation, run waves. A wait-graph watches every conversation and breaks loops before they spin.
Everything you just saw runs on your machine.
VOICE
Talk to it. It talks back.
Hold the dictation hotkey (Control+Shift by default) and speak, and nala answers in one of 15 on-device voices. Speech to text and text to speech both run entirely on your machine: no cloud round-trip, no API keys, no per-word bill for talking to your own agent.
It doesn't stop at the terminal: one shortcut dictates into any window, from a browser tab to a doc to an email, and session recaps land in the workspace's running notes. A router and a voice in the same process are the raw parts for building voice agents of your own, and prototyping one costs exactly nothing. The models are already on your machine.
LOCAL TEXT TO SPEECHNO API KEYON-DEVICE STT
THE BASE LAYER
The terminal was the
first operating system.nala is its agentic
evolution.
From a terminal you can already do anything on a machine. nala keeps that reach and puts a fleet of agents behind it. Speak or type, and conduct the work. Not a plugin sitting on top of your tools: a command center for the agentic era, and how you and your agents run the machine.
“A few days of living in it and I stopped toggling between terminals and virtual desktops. I command whole teams of agents, and put agents in charge of other agents. It feels like superpowers.”
Agent-nativeTerminal-nativeVoice in & outLocal-firstDurable stateYours to extend
SOFTWARE FACTORIESTen agents in one repo
Ten agents in one repo
is not a team. It is a merge conflict.
Real teams do not collaborate by typing into the same file at the same time. They converge on shared work: a spec, a plan, a review, a decision. nala gives agents that same substrate. They talk over a protocol and write to versioned artifacts, so the work composes instead of colliding.
Artifacts are the shared work product
Specs, plans, stories, tickets, reviews, decisions and runbooks. Each one versioned, hashed, assignable to an agent, and linked to its parent. Agents produce documents that stack, not simultaneous edits to one checkout.
Every artifact knows where it came from
Which agent wrote it, on which model and provider, from which inputs, at which revision. When ten agents contribute to one plan you can still reconstruct who decided what.
Delivery you can audit
A send reports what actually arrived. When nala cannot observe a submission it says unknown instead of reporting success, which is the failure that quietly breaks every other multi-agent setup.
Any model, one protocol
Claude, Codex, Grok and Gemini work the same artifacts side by side in one workspace. Your process is not a bet on a single vendor.
We are not going to quote you a throughput multiple. nala is new and we publish no invented numbers. Here is exactly how it works. Run it and measure your own.
WHY NALA
The first wave bolted agents in. nala starts from the fleet.
AI editors kept the file at the center and parked agents in a chat panel. Agent CLIs run one session at a time with no shared state. nala inverts both.
Where agents live today
- A text editor with a chat panel; the file stays the center
- One session at a time, no shared state between agents
- Context dies when the window closes
- Tied to one model, one editor, one vendor
nala
- The fleet is the center: tasks, terminals, and approvals are first-class
- Agents share a task board and hand work off durably
- Sessions survive a reboot, so context compounds
- Provider-neutral: the agents you already pay for run better inside nala
The same fleet, everywhere you work
One install. A desktop command center, a resident terminal agent, and a CLI, all conducting the same fleet.
PRICING
Free. Bring your own agents.
The whole product is free today: extend the subscriptions you already pay for, with voice and orchestration included. No metering, no markup.
$0 today
The whole app. Not a trial, not crippleware.
nala: free preview
Download nala, connect supported tools you already use, and run the local-first desktop app and agent without entering a credit card. nala does not currently sell a hosted model service, credits, subscriptions, or team plan.
FAQ
Questions, answered
Straight answers about the product, the free preview, and what runs on your machine.
nala is an agentic IDE: a private, local-first workspace where your AI coding agents run side by side. Connect Claude Code, Codex, Devin, Cursor, Grok, Kimi, or any custom CLI agent, and nala handles durable tasks, artifact sharing, agent-to-agent messaging, voice, and MCP, with no cloud host required.
WHAT'S NEXT
The base layer is free. Take it.
Bring your own models and agents. Durable terminals, agent-to-agent orchestration, on-device voice, and MCP, all local-first, no cloud host required. Your machine is the empire.
The Nerve Center for Agentic Engineering
Start building with nala today
Download nala free and be among the first to orchestrate your AI coding agents. Product updates, tips, and early-access notes in your inbox.
