AgentScope Studio
A local visualization toolkit for AgentScope developers, making development, debugging and evaluation of agent applications transparent, simple and fun.
- Source repo
- agentscope-ai/agentscope-studio
- Stars
- ★ 674
- Last updated
- 2mo ago
- License
- Apache-2.0
- Primary language
- TypeScript
- FA score
- 53/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platform
- Cost
- Free, no paid service needed
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- AgentScope developers who want to watch agent conversations and behavior in real time through a local web interface.
- Not a fit if
- New projects needing ongoing maintenance: the repo is archived in favor of the built-in Web UI in AgentScope 2.0
- Teams building agents outside the AgentScope framework
- Website
- agentscope-ai.github.io ↗ set on GitHub
- Maintenance
- Commits in the last 6 monthslast commit 2026-09-02
- Source review
- 53/100 · Major gaps
What does this agent do, and when should you use it?
AgentScope Studio is a locally run visualization toolkit (agentscope-ai/agentscope-studio, Apache-2.0 licensed) for developers working with the AgentScope framework. It provides project management (Projects and Runs), a chatbot-style real-time runtime interface, OpenTelemetry-based trace visualization for LLM calls, token usage and agent invocations, statistically oriented agent evaluation, and a built-in Copilot named Friday. AgentScope applications connect by setting the studio_url field in agentscope.init. It is installed globally via npm and launched with the as_studio command, then accessed at localhost:3000; source and Docker deployment paths are also documented. Note that the repository carries an archive notice: it is no longer maintained as a standalone tool, users are directed to AgentScope 2.0's built-in Web UI, and this repository will become read-only and be archived soon.
Studio runs a local web UI (default http://localhost:3000). Once an AgentScope application registers via agentscope.init(studio_url="http://localhost:3000"), Studio surfaces its activity: Projects/Runs organize and manage AgentScope projects; a chatbot-style interface enables real-time agent interaction; OpenTelemetry-based tracing visualizes LLM calls, token usage and agent invocations; an evaluation module analyzes agent performance from a statistical perspective; and the built-in Copilot Friday offers a development assistant, a playground for rapid secondary development, and an integration hub. Installation paths are npm install -g @agentscope/studio followed by as_studio, or cloning the source and running npm install && npm run dev, with optional Docker deployment documented in docker/README.md.
- AgentScope developers who want to watch agent conversations and behavior in real time through a local web interface.
- Engineers debugging LLM call chains and token consumption via the OpenTelemetry trace views.
- Teams building multi-agent applications who need Projects and Runs to organize and manage experiments.
- Researchers or evaluators who need statistically oriented analysis of agent performance.
- Developers prototyping secondary development quickly in the playground using the built-in Copilot Friday.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npm install -g @agentscope/studio
- Runtime
- Node.js 20.0.0+
- Install notes cover
- Docker
How do you install or deploy this agent?
Prerequisites: Node.js >= 20.0.0 and npm >= 10.0.0 (with nvm, run nvm use to switch automatically). Recommended install from npm:
npm install -g @agentscope/studio
# Start AgentScope Studio
as_studioOr run from source:
git clone https://github.com/agentscope-ai/agentscope-studio
cd agentscope-studio
npm install
# Start in development mode
npm run devDocker deployment is documented in the repository's docker/README.md.
How do you use this agent?
After starting Studio (default http://localhost:3000), connect your AgentScope application by setting studio_url in the initializer:
import agentscope
agentscope.init(
# ...
studio_url="http://localhost:3000"
)
# ...Once connected, use the browser UI to manage Projects/Runs, interact with agents in real time through the chat interface, and inspect OpenTelemetry traces and evaluation analysis.
What are this agent's strengths and limitations?
- Native integration with the AgentScope framework — a single studio_url setting wires up runtime visualization.
- OpenTelemetry-based tracing gives auditable views of LLM calls, token usage and agent invocation chains.
- Runs locally under Apache-2.0 with three documented deployment paths (npm, source, Docker).
- An archive notice states it is no longer maintained as a standalone tool and points users to AgentScope 2.0 — a real migration risk.
- Tightly coupled to AgentScope: applications not built on AgentScope have no documented way to connect.
- Requires a local runtime of Node.js >= 20.0.0 and npm >= 10.0.0.
How does this agent compare with similar options?
The README names its successor: the built-in Web UI shipped with AgentScope 2.0 (a FastAPI backend paired with a pre-built frontend in examples/web_ui), which adds multi-session isolation, Agent Team (leader–worker) orchestration with task planning, permission and human-in-the-loop control, IM channels (Feishu/Discord), MCP and Skill Hub management, session persistence and task scheduling.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| AgentScope Studio This agent | 53 · Major gaps | Web appFree | ★ 674 | 2mo ago | TypeScript | — |
| Agent Skills for Context Engineering | 66 · Some gaps | Agent plugin / skillFree + model costs | ★ 18k | 1d ago | Python | Codex · Claude Code |
| NVIDIA NeMo Agent Toolkit | 61 · Some gaps | CLIFree + model costs | ★ 2.7k | 2d ago | Python | — |
| Judgeval | 53 · Major gaps | Library / SDKFree + model costs | ★ 1.1k | 2d ago | Python | OpenAI API |
How does FollowAgents rate this agent?
Why each dimension lost points
Friday registers execute_python_code and execute_shell_command by default; write_text_file requires explicit --writePermission, showing partial least privilege. However, shell/python execution is extremely powerful and permission control rests only on the sys prompt ('request permission, prohibit catastrophic operations') with no technical confirmation gate. Data flow (local JSON session persistence, studio_url forwarding) is visible in code (2), but API keys pass via CLI args with no redaction guidance; sensitive-data handling is thin. No rollback mechanism; sessions are append-only. Dependencies are version-pinned but no lockfile/audit evidence. Attribution is clear (author, team, Apache-2.0).
Internally consistent: dual interactive/non-interactive paths for port conflicts, gRPC failure degrades to HTTP /v1/traces, graceful SIGINT/SIGTERM shutdown; JSONSession try/except handles API changes; error messages are specific. Dependencies are mainstream, maintained npm packages with reasonable constraints.
Audience (AgentScope developers) and scenarios are clear; capability boundaries (visualization/tracing/evaluation/Friday) are well described. Tool trigger precision relies on prompt text with no structured trigger definitions; max_iters=50 in code vs max_turns=20 in the prompt is a minor inconsistency.
Good information architecture (README, docs tree, docker notes), install notes with version checks and nvm tip, stable naming (@agentscope/studio, as_studio), CI verifies package.json/release tag version consistency. Full Apache-2.0 license text earns 3. But no FAQ/example detail docs (links only), no known-limitations section, and the repo is announced as archived and unmaintained — maintenance responsibility is materially absent, only a migration pointer remains.
Output usability is decent (visualization, tracing, health-check scripts). Marginal value as a local dev visualization tool for AgentScope users exists, but the project is deprecated and superseded by AgentScope 2.0's built-in Web UI, sharply reducing cost-benefit.
Claims are traceable: feature list maps to trpc/otel/socket modules in code; arXiv paper, npm badge, and CI workflows corroborate; facts vs inference are separated in the README (archive notice is explicit). Marketing claims ('powerful') lack test evidence, and the packet contains no test code.
- The repo is announced as archived and read-only; no future security fixes — migrate to AgentScope 2.0.
- Friday can execute arbitrary Python and shell commands by default; permission control is prompt-level only. Run in an isolated environment and grant writePermission cautiously.
- API keys are passed as CLI arguments and may leak into shell history or process listings; handle with care.
- No known-vulnerability disclosure, dependency audit, or lockfile evidence; dependency security is unverified.
FAQ
Can I still use AgentScope Studio?
Does it cost anything?
Can I use Studio without the AgentScope framework?
What runtime do I need?
How is it deployed?
Is AgentScope Studio open source?
Is AgentScope Studio safe to use?
Is AgentScope Studio still maintained?
Who makes AgentScope Studio?
What is AgentScope Studio's official website?
Thanks — we use this to improve the page.