NeuroSquad Terminal Dashboard
Run and manage several AI coding CLIs from one terminal dashboard.
- Source repo
- glmn-ai/neurosquad-cli
- Stars
- ★ 53
- Last updated
- 5d ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 76/100 · Good
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude CodeOpenAI API · Claude API (Partial support)
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Developers handling several coding tasks in parallel can run Claude Code, Codex, or OpenCode sessions and isolate changes with worktrees.
- Not a fit if
- Teams that require stable 1.0 behavior
- Users who need voice dictation on Windows ARM64
- Source review
- 76/100 · Good
What does this agent do, and when should you use it?
nsq is a Node.js CLI that starts Claude Code, Codex, OpenCode, or any command and manages its terminal session through a per-user background daemon. Its terminal dashboard brings workspaces and live terminals into a grid, with full-screen access, prompt sending, permission responses, and status display. Supported CLIs use their own hooks to report states such as “needs you” along with the question text; arbitrary commands report only working or finished. The tool also offers per-agent Git worktrees, cost views, notifications, provider configuration, experimental phone access, and local voice dictation. It runs on the machine where nsq is installed and supports Windows, macOS, Linux, and SSH workflows; version 0.1.0 is still a preview.
From a project directory, run commands such as nsq run claude "fix the flaky checkout test" --worktree to start a CLI, optionally giving that agent its own branch and worktree; nsq run -- <command…> starts an arbitrary command. A per-user daemon owns the terminal sessions, so work continues after the dashboard closes and can be reconnected to later with nsq. The dashboard displays workspaces, terminal output, and status; users can open an agent full screen, send prompts, interrupt work, and answer permission requests. nsq ls, nsq answer, nsq diff, and nsq cost provide scriptable management. Users can also configure OpenRouter or their own model servers with OpenAI- or Anthropic-compatible APIs, and inspect approval logs and notifications.
- Developers handling several coding tasks in parallel can run Claude Code, Codex, or OpenCode sessions and isolate changes with worktrees.
- Engineers working over SSH or on a remote Linux server can let the daemon keep agents running after their terminal disconnects.
- A developer supervising permission requests can read the actual question and approve, always allow, or reject it in the dashboard.
- Developers using local or self-hosted models can add Ollama, LM Studio, llama.cpp, or a compatible API server and select its models.
- Users who need to check agent status or respond from a phone can opt into experimental pairing and LAN or Cloudflare tunnel access.
How do you install or deploy this agent?
Requires Node.js 22.13 or newer. Install globally with npm or try it without installing:
npm install -g neurosquad
nsqYou can also run it without installing:
npx neurosquadThe README also lists Homebrew, Scoop, and install scripts, but does not give their commands here. No API key is required to run nsq; configure credentials only when using OpenRouter or a self-hosted model server that requires them.
How do you use this agent?
Start a Claude Code session in its own worktree, launch Codex, and open the dashboard:
cd ~/code/my-app
nsq run claude "fix the flaky checkout test" --worktree
nsq run codex --name reviewer
nsqUse the arrow keys to select an agent, Enter to open it full screen, and Ctrl+] to return. Press y, a, or n to answer a permission request; q closes the dashboard while agents keep running. You can also manage sessions from the CLI:
nsq ls --json
nsq send reviewer "Review the checkout changes"
nsq answer reviewer yes
nsq diff reviewer
nsq costTo use OpenRouter, set a key and select a provider and model:
nsq openrouter set-key
nsq run claude --provider openrouter --model anthropic/claude-sonnet-4.5What are this agent's strengths and limitations?
- A single dashboard manages live terminals for multiple CLIs, while the daemon keeps sessions running after the dashboard closes.
- Native hooks from supported harnesses report waiting states and show the permission question text.
--worktreegives each agent a separate branch and checkout to reduce conflicts between parallel changes.- Supports OpenRouter and several user-provided model servers, while also allowing a CLI's own login.
- Includes local speech recognition and an approvals log that the README says stays on the machine.
- Version 0.1.0 is a preview; commands, keys, and configuration may change before 1.0.
- Requires Node.js 22.13 or newer; users also need the chosen coding CLI installed for its harness-specific hooks.
- Arbitrary commands lack harness hooks, so nsq reports only working or finished for them, not “needs you.”
- Phone access is experimental; online remote access uses a Cloudflare tunnel and must be explicitly enabled.
- Voice dictation is unavailable on Windows ARM64, and first use downloads a model of about 670 MB.
How does this agent compare with similar options?
The README distinguishes nsq from the NeuroSquad desktop app: nsq is aimed at people who prefer a terminal, work over SSH or on Linux servers, or want a small open tool; the desktop app offers a canvas, 20+ CLIs, agent-to-agent arrows, and browser, notes, and plugin cards.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| NeuroSquad Terminal Dashboard This agent | 76 · Good | CLIFree + model costs | ★ 53 | 5d ago | TypeScript | Codex · Claude Code |
| Tessera | 74 · Some gaps | Desktop appFree + model costs | ★ 315 | today | TypeScript | Codex · Claude Code |
| OpenCode Telegram Bot | 64 · Some gaps | CLIFree + model costs | ★ 1.3k | 1d ago | TypeScript | — |
| Plannotator | 78 · Good | CLIFreemium | ★ 9.5k | today | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The README says keys go in the OS keyring, approval logs stay local and mask secret-looking content, phone controls are limited, and remote access must be explicitly enabled. Least privilege, confirmation, and external effects score 2: arbitrary commands and the agents themselves can still perform consequential actions, and automatic updates are enabled by default. Data-flow transparency scores 3 because the docs enumerate network requests, log contents, and phone push contents. Sensitive-data handling scores 2: protections are described, but the supplied material does not show their implementation. Dependency security scores 2: CI actions are pinned to commits and some downloads are hash checked, but routine dependency auditing is not shown. Rollback scores 2: stopping, removal, and safe update switching are documented, but no general recovery procedure is supplied. Source attribution scores 2: the license names an author and contributors and a private reporting channel is provided, but publisher identity is unverified.
The docs, package metadata, and workflows broadly agree on Node requirements, platforms, and preview status, so self-consistency scores 2; the material does not show actual run results. Dependency availability scores 2: Node requirements, prebuilt platforms, and supported agents are listed, but external CLIs, network access, and native components remain prerequisites. Failure messages score 1: troubleshooting docs and a doctor command are described, but error-handling implementation or message examples are not provided.
Audience and scenarios score 3: the material covers terminal use, SSH, parallel agents, arbitrary commands, and several model servers. Capability boundaries score 3: it explains that commands without hooks do not report waiting status, that phones cannot start agents or type arbitrary keys, and that phone support is experimental. Trigger precision scores 2: status mainly comes from CLI hooks, while a five-minute quiet-screen heuristic identifies possible stalls and can produce false positives. Environment fit scores 3: Windows, macOS, Linux, SSH, Node requirements, and platform exceptions are documented.
Information architecture scores 3: the README clearly organizes installation, quick start, commands, privacy, and topic guides. Install notes score 3: npm, npx, Node requirements, and pointers to other installation methods are provided. Naming stability scores 2: nsq and neurosquad-cli are generally distinguishable, but the preview notice warns that commands, keys, and configuration may change. Examples and FAQ score 3: there are extensive runnable command examples and a troubleshooting entry point. Known limitations score 2: preview status, experimental features, platform exceptions, and status-detection boundaries are disclosed, but no comprehensive consolidated limitations list is shown. License scores 3: the full MIT text is supplied and matches the metadata. Versioning and changelog score 1: 0.1.0 preview and release-process links are given, but no actual changelog appears in the supplied material. Maintenance responsibility scores 2: contribution, security-reporting, supported-version, and update paths are described, but publisher identity is unverified.
Output usability scores 3: the dashboard presents agent status, questions, costs, and approval records, with scripting interfaces also documented. Marginal value scores 3: it brings long-running agents, status, and interaction into one place, and the README explains the benefits over running CLIs separately. Cost-benefit scores 2: the software is open source and free, but multiple agents, speech-model downloads, and a background daemon have resource and maintenance costs that the material does not quantify.
Claim traceability scores 2: most product and privacy claims point to specific guides, and CI and live end-to-end workflow descriptions state their intended coverage; these materials do not prove actual passing runs or implementation outcomes. Cross-source corroboration scores 2: the README, package metadata, license, and workflows support some basic facts, but do not independently verify core behavior. Fact-inference separation scores 2: the docs label experimental features, preview status, and heuristic detection, though many performance and security assurances remain project claims.
- This is a 0.1.0 preview; the project says commands, keys, and configuration may change.
- Phone access can control agents. Online mode is opt-in, and pairing links and tokens should be treated like passwords.
- Quiet detection is heuristic; the README also says some prompts cannot be identified precisely through hooks.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Minor Changes - [#76](https://github.com/glmn-ai/neurosquad-cli/pull/76) [7383db9](https://github.com/glmn-ai/neurosquad-cli/commit/7383db901e93ed65c604295157adf9560d8d6b40) Thanks [@neurosquad-dev-bot](https://github.com/apps/neurosquad-dev-bot)! - The dashboard selects text and opens links itself. It holds the mouse, so the terminal couldn't do either: now a drag over an agent's text copies it (OSC 52, plus the system clipboard when local) and shows it inverted, and a click on an http(s) link opens it in the browser (copied instead over SSH). term-view gains linkAt, selectionText, …View on GitHub ↗ -
Release notes excerpt
Minor Changes - [#76](https://github.com/glmn-ai/neurosquad-cli/pull/76) [7383db9](https://github.com/glmn-ai/neurosquad-cli/commit/7383db901e93ed65c604295157adf9560d8d6b40) Thanks [@neurosquad-dev-bot](https://github.com/apps/neurosquad-dev-bot)! - The dashboard selects text and opens links itself. It holds the mouse, so the terminal couldn't do either: now a drag over an agent's text copies it (OSC 52, plus the system clipboard when local) and shows it inverted, and a click on an http(s) link opens it in the browser (copied instead over SSH). term-view gains linkAt, selectionText, …View on GitHub ↗ -
Release notes excerpt
Patch Changes - [#74](https://github.com/glmn-ai/neurosquad-cli/pull/74) [b70076e](https://github.com/glmn-ai/neurosquad-cli/commit/b70076e9d305f47b61273f5e2b10aaeca7a831fb) Thanks [@neurosquad-dev-bot](https://github.com/apps/neurosquad-dev-bot)! - Prices GPT-6 Astra, GPT-6 Sol, GPT-6.1 Sol and GPT-6 Luna (OpenAI's list prices, with their rates above 272K prompt tokens). Codex's codex-auto-review and gpt-5.3-codex-spark have no published price and stay "no price". The price table is now dated 2026-10-13. …View on GitHub ↗
FAQ
Do I need a paid model account to run nsq?
Will agents stop when I close the terminal or dashboard?
nsq down stops the daemon and its agents.Can nsq detect when every command needs my approval?
Is phone access enabled by default?
nsq phone on and pair a device; access from outside your network also requires the explicit --online option.