Dev & Engineering mcp-clientknowledge-graphlive2dvoice-interfaceopenclawmulti-agentstreaming-tool-calls

NagaAgent

An anime-styled all-in-one AI assistant with streaming tool calls, knowledge-graph memory, Live2D avatars, voice interaction, and OpenClaw integration.

FollowAgents review · FARS-2.1
Not recommended
45/ 100 5-point scale 2.3 / 5
1 2 3 4 5 6
1Trust8 / 29 · 1.4/5

Evidence shows: README mentions placeholder API key rejection, preserving real keys on save, and Skill path traversal rejection, but lacks a full permission model or user confirmation mechanism. Deductions: no explicit user confirmation flow, limited data flow transparency, insufficient sensitive data handling details, un-audited dependency security, external effects (e.g., OpenClaw auto-execution) without user confirmation, no rollback mechanism, source attribution only author info.

2Reliability8 / 14 · 2.9/5

Evidence shows: README and test files demonstrate config path priority, tool execution, route tests, showing good self-consistency. Deductions: dependency availability unverified (e.g., Neo4j, OpenClaw), failure messages partial coverage.

3Adaptability9 / 18 · 2.5/5

Evidence shows: README describes multiple use cases (chat, memory, game guides), supports Windows/macOS/Linux, flexible config. Deductions: capability boundaries unclear (e.g., OpenClaw auto-execution scope), trigger precision (e.g., tool call parsing) described but not fully verified, environment fit details (e.g., packaging) not deep.

4Convention10 / 18 · 2.8/5

Evidence shows: README well-structured, install steps detailed, naming stable (version unified), examples and FAQ (troubleshooting), license clear (AGPL-3.0 dual), changelog detailed. Deductions: known limitations not explicitly listed, maintenance responsibility only author info, no contributor or maintenance policy.

5Effectiveness7 / 13 · 2.7/5

Evidence shows: output usability (e.g., structured tool results) described, marginal value (e.g., multi-agent collaboration) evident. Deductions: cost-benefit not quantified, no resource or performance data.

6Verifiability3 / 8 · 1.9/5

Evidence shows: README feature descriptions partially correspond to code structure (e.g., directory), tests provide some verification. Deductions: claim traceability insufficient (e.g., feature claims not fully mapped to code), cross-source corroboration limited (only within repo), fact-inference separation unclear.

Evidence confidence: Low Reviewed Aug 11, 2026 Reviewed revision c2caa9079b9e
Safety controls not found in source: rollback or recovery path
Before you use it
  • The repository includes auto-execution tools (e.g., OpenClaw) and web exploration features that may have external effects; configure permissions and user confirmation carefully.
  • Dependencies are numerous and not security-audited; check for vulnerabilities before use.
  • License is AGPL-3.0 dual-licensed; commercial use requires attention to closed-source license terms.
Review evidence [1][2][3][4][5][6][7][8]
See the full review method →

What does this agent do, and when should you use it?

NagaAgent is a full-stack agent framework for personal assistants, built with an Electron + Vue 3 frontend and multiple Python microservice backends. It features streaming tool calls (parsed from inline tool code blocks, not dependent on OpenAI Function Calling), GRAG knowledge-graph memory (Neo4j or local), Live2D virtual avatar interaction, TTS/ASR voice interaction, and OpenClaw computer control. The software comprises five independent services: API Server (port 8000), Agent Server (8001), MCP Server (8003), Memory Server (8004), and Voice Service (5048), orchestrated by main.py. The main panel offers eight modules: Chat, MindSea (knowledge graph visualization), Skill Workshop, Naga Network (forum), Cardinal Market, Terminal Settings, Music Studio, and Floating Orb. It uses a dual license: AGPL-3.0 for open source and a proprietary license for commercial use, supporting Windows/macOS/Linux.

NagaAgent performs the following concrete operations: 1) Streams dialogue via SSE, parses JSON tool calls from inline ``tool`` code blocks, and routes them to MCP Manager or OpenClaw, supporting up to 5 tool loops. 2) Extracts quintuples (subject, predicate, object, etc.) from conversation and stores them in Neo4j or local JSON, injecting retrieved context via RAG. 3) Renders Live2D models with a 4-channel orthogonal animation system (state, action, emotion, tracking) and SSAA anti-aliasing. 4) Provides TTS (Edge-TTS) and ASR (FunASR) voice interaction, including real-time full-duplex voice chat (Qwen Omni). 5) Controls the computer via OpenClaw Gateway, with task scheduling and memory compression. 6) Includes 9 built-in MCP agents (weather, launcher, game guide, search, crawler, browser automation, vision, MQTT, document). 7) Offers community forum and credit/affinity system.

  1. Individual users seeking a desktop companion AI with anime avatar for friendly chat and tool use like weather queries and app launching.
  2. Gamers using the game guide agent to automatically recognize game screenshots and receive suggestions, optionally automating gameplay via MAA plugins.
  3. Developers and power users integrating OpenClaw to execute local computer tasks through natural language, such as file operations and browser automation.
  4. Users needing knowledge management, leveraging Neo4j graph memory to record facts and automatically retrieve them in future conversations.
  5. Content creators using voice interaction for hands-free dialogue or TTS to generate audio content.
  6. Community members engaging in forum activities, earning credits and affinity through daily check-ins to unlock model quotas or music albums.

What are this agent's strengths and limitations?

Pros
  • Streaming tool calls are provider-agnostic, working with any OpenAI-compatible API and also supporting Anthropic format, reducing vendor lock-in.
  • Rich built-in components including Live2D avatar, voice interaction, and game guide, offering a unique anime-style experience.
  • Knowledge-graph memory (GRAG) supports local Neo4j or cloud NagaMemory, enhancing context understanding.
  • Includes MCP server and skill workshop with community skill installation, enabling extensibility.
Limitations
  • Desktop-focused, relying on Electron and multiple Python microservices, making deployment and configuration complex.
  • Core services like NagaBusiness gateway and NagaMemory require login and may incur credits; functionality is limited when not logged in.
  • Dual licensing (AGPL-3.0 + commercial) may impose additional requirements for commercial use.
  • Real-time voice chat depends on DashScope (Qwen) API, introducing specific vendor dependency.

How do you install or deploy this agent?

Requirements: Python 3.11 (>=3.11,<3.12), optional uv, optional Neo4j. Steps: 1) git clone https://github.com/Xxiii8322766509/NagaAgent.git; 2) cd NagaAgent; 3) Install frontend: cd frontend && npm install && cd..; 4) Install backend (option A: uv sync; option B: python -m venv .venv && source .venv/bin/activate (Windows: .\venv\Scripts\activate) && pip install -r requirements.txt). 5) Copy config.json.example to config.json and fill in LLM API details (api_key, base_url, model).

How do you use this agent?

After configuration, run cd frontend && npm run dev to launch the app (one-click startup supported). Log in to auto-configure API keys, or choose 'Don't log in, use open-source version' and set local keys manually. From the main panel, access Chat, MindSea, Skill Workshop, etc. Configure model connection, memory, and audio-visual settings in Terminal Settings. To use OpenClaw, ensure the Agent Server is running and enabled in settings.

FAQ

How do I configure multiple models?
Supports any OpenAI-compatible API (DeepSeek, Qwen, etc.) and Anthropic format. Set api_format to "openai" or "anthropic" in config.json, along with the corresponding api_key and base_url.
What is required to use knowledge graph memory?
Optionally install Neo4j and configure the grag section; without it, it stores locally in JSON and functionality is unaffected.
How can I customize Live2D models?
Upload a complete model directory (including .model3.json etc.) in Terminal Settings -> Audio-Visual Settings -> Live2D Model, or define a custom role in Cardinal Market -> Character Registration.
How does OpenClaw integration work?
It sends commands via OpenClaw Gateway (port 20789), supports automatic installation and configuration, and allows creating exploration tasks with monitoring.
What if startup gets stuck on the progress bar?
Check that your API key is correct, wait for a restart prompt after 3 seconds, or run python main.py --check-env --force-check for diagnostics.

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related agents