Naraya CLI

One sign-in, every model. A coding agent that delegates specialist work to focused subagents and brings evidence to every claim.

Naraya CLI is an AI coding agent that:

• Routes all model traffic through router.naraya.ai with one API key — Claude, GPT, Gemini, DeepSeek, GLM, Kimi, MiniMax, Mistral, and more.
• Ships a roster of focused subagents (nara-build, nara-architect, nara-release, nara-review, nara-debug, nara-plan, nara-explore, nara-search, nara-fe, nara-droid) — each with its own skills and references.
• Provides a delegate tool that fans work out to subagents in parallel, sequential chains, or single focus — each in an isolated context window.

Table of contents

• Install
• Quickstart
• Usage
• Subagents
• Skills
• Support
• License

Install

# Requires Node.js >= 22.19
npm install -g @naraya/cli

Then sign in with your Naraya account:

naraya login

Naraya CLI is proprietary software. Access requires a valid Naraya account and API key — sign up at router.naraya.ai.

Quickstart

# 1. Sign in (one-time)
naraya login

# 2. Verify auth
naraya status

# 3. Launch the agent in the current project
naraya

Inside the agent, press / for the command palette, /agents to list subagents, /skills to list skills.

# Quick subcommands without launching the TUI
naraya login              # sign in to router.naraya.ai
naraya logout             # wipe local credentials
naraya status             # show auth + quota + current model
naraya status --usd       # same, with USD cost equivalent

Usage

Interactive mode

naraya                        # launch in cwd
naraya --model naraya/glm-5   # override default model for this session
naraya --continue             # resume the last session

Slash commands inside the TUI:

| Command | Effect | |---|---| | /agents | list subagents | | /agent-model <name> | change a subagent's model | | /skill:<name> [args] | load a skill | | /skills | list available skills | | /help | full command palette |

Headless / scripted

naraya -p "audit the auth module for OWASP top 10"   # single prompt, prints result, exits
naraya -p --json <task>                              # machine-readable output

Delegation patterns

The agent delegates to specialists via the delegate tool. You do not invoke it directly — you phrase the request and the model decides.

// One focused specialist
delegate { agent: "nara-droid", task: "..." }

// Independent units in parallel
delegate { tasks: [
  { agent: "nara-search",  task: "research X" },
  { agent: "nara-explore", task: "find usages of Y" },
  { agent: "nara-plan",    task: "draft plan for Z" }
]}

// Sequential, {previous} injects the prior step's output
delegate { chain: [
  { agent: "nara-explore", task: "map the auth flow" },
  { agent: "nara-plan",    task: "draft a plan based on {previous}" },
  { agent: "nara-review",  task: "review the plan in {previous}" }
]}

Subagents

Naraya ships a roster of focused subagents. The model picks based on the unit of work, not the surface form.

Coordination

| Agent | Purpose | When to use | |---|---|---| | nara-build | Implements features, bugfixes, refactors, releases | Top-level orchestrator. Default for "implement X", "fix Y", "refactor Z" | | nara-architect | System design, ADRs, component decomposition | Greenfield design, scaling decisions, API contracts, trade-offs. Read-only | | nara-release | Version sync, verification, deploy safety | Before any production push, store release, or schema migration | | nara-review | Multi-aspect code review (security, perf, a11y, API, tests) | Pre-merge review, audit, self-review. Does not fix | | nara-debug | 4-phase root-cause debugging | "X is broken", flaky tests, intermittent failures |

Investigation (read-only)

| Agent | Purpose | When to use | |---|---|---| | nara-plan | Implementation plans, never edits code | Pair with nara-build for hand-off | | nara-explore | Fast codebase navigation | "Where is X defined?", "who calls Y?", quick map | | nara-search | Evidence-first research with citations | Library comparison, version migration, doc lookup |

Specialist (writes code in their domain)

| Agent | Purpose | When to use | |---|---|---| | nara-fe | Frontend UI/UX (React, Vue, Svelte, Angular, CSS, Tailwind, a11y) | UI changes, component design, responsive, accessibility | | nara-droid | Native Android (Kotlin, Gradle, Compose, Room, Hilt, adb, APK/AAB) | Android tasks, build failures, release |

If multiple roles apply, the model fans out to parallel subagents.

Skills

Naraya bundles a large pool of skills that any agent can load on demand via /skill:<name>. Highlights:

| Skill | What it covers | |---|---| | systematic-debugging | 4-phase root-cause debugging | | test-driven-development | RED-GREEN-REFACTOR discipline | | writing-plans | Implementation plans with bite-sized tasks | | security, auth-identity, compliance-governance | Security review and compliance | | architecture, api-design-patterns, distributed-systems | System design | | android-kotlin, compose-patterns, gradle-troubleshoot | Android development | | react-patterns, css-patterns | Frontend |

Skills vs subagents: a skill is a workflow / lens the agent applies inline. A subagent is a separate context that runs in parallel. Use skills for "do X the right way", subagents for "do X in isolation".

For project-specific guidance, add an AGENTS.md to your repository — the agent reads it automatically.

Support

• Account, billing, API keys: router.naraya.ai
• Sign-in issues: run naraya logout then naraya login to refresh your session.

License

Proprietary. Copyright © 2026 PT. Naraya Teknologi Indonesia. All rights reserved. See LICENSE.