← Open Source
grp06

openclaw-studio

A clean web dashboard for OpenClaw. Connect your Gateway, manage agents, and ship faster. ⭐️ Star if you like it.

AI EngineeringBuild agent UITypeScript
Open on GitHub
Momentum
+1stars in 24 hours+0.0%
2.03k
Stars
294
Forks
+-2
This week
4
Contributors
Created 2026-01-28 · Updated 2026-10-15 · #5204 today
Top developers
grp06jfedoroelor

grp06/openclaw-studio is ranked 5,204 with 2,034 stars.

Same-category comparison

grp06/openclaw-studio is ranked 5,204 among these 8 peers; nottelabs/notte is ranked 5,202.

Same-category comparison
NameRankMetrics
Cloud-CV/EvalAI5,194
Language:
Python
Stars:
2,047
Forks:
983
Description:
:cloud: :rocket: :bar_chart: :chart_with_upwards_trend: Evaluating state of the art in AI.
zereight/gitlab-mcp5,197
Language:
TypeScript
Stars:
2,045
Forks:
358
Description:
MCP server for GitLab: 100+ tools for projects, MRs, issues, pipelines, wikis and more via the GitLab REST/GraphQL API.
KudoAI/chatgpt.js5,198
Language:
JavaScript
Stars:
2,042
Forks:
178
Description:
🤖 A powerful, lightweight JavaScript library + interactive CLI for ChatGPT.
nottelabs/notte5,202
Language:
Python
Stars:
2,039
Forks:
186
Description:
Cloud browser infrastructure and web automation platform for your AI and coding agents.
msoedov/agentic_security5,207
Language:
Python
Stars:
2,022
Forks:
296
Description:
Agentic LLM Vulnerability Scanner / AI red teaming kit 🧪.
tatsu-lab/alpaca_eval5,208
Language:
Jupyter Notebook
Stars:
2,021
Forks:
314
Description:
An automatic evaluator for instruction-following language models. Human-validated, high-quality, cheap, and fast.
doobidoo/mcp-memory-service5,209
Language:
Python
Stars:
2,019
Forks:
350
Description:
Open-source persistent memory for AI agent pipelines (LangGraph, CrewAI, AutoGen) and Claude. REST API + knowledge graph + autonomous consolidation.
shcherbak-ai/contextgem5,212
Language:
Python
Stars:
2,010
Forks:
185
Description:
Extract structured data and insights from documents using LLMs and decision models.
Key parameters and specifications
Key parameters and specifications
ParameterValue
LanguageTypeScript
CategoryAI Engineering
SubcategoryBuild agent UI
KeywordsAgent-ready APIs, Client integration, Webdev
Stars2,034
Forks294
Contributors4
Created date2026-01-28
Updated date2026-10-15
Stars in 1 day1
Stars in 7 days-2
Rank5,204
Previous rank18,581
Rank change13,377
7-day stars rank18,251
Previous 7-day stars rank18,205
7-day stars rank change-46
Total stars rank6,269
Previous total stars rank6,261
Total stars rank change-8
CountryUSA
README summary

OpenClaw Studio OpenClaw Studio is a clean web dashboard for OpenClaw. Use it to connect to your Gateway, see your agents, chat, manage approvals, and configure jobs from one place. ⭐ It helps more developers discover the project. Get Started (Pick Your Setup) If your Gateway is already running, pick the scenario that matches where your Gateway and Studio will run: A. Gateway local, Studio local (same computer) B. Gateway in the cloud, Studio local (your laptop) C. Studio in the

cloud, Gateway in the cloud All setups use the same install/run path (recommended): Two links matter: 1. Studio -> Gateway always means "the Studio host. " If Studio and OpenClaw share a machine, the upstream should usually stay at even when that machine is a cloud VM. 9+ (LTS recommended) An OpenClaw Gateway URL + token, or a local OpenClaw install Studio can detect Tailscale (optional, recommended for remote access) A) Gateway local, Studio local (same computer) 1. In Studio, set: Upstream

URL: Upstream Token: your gateway token (for example: ) B) Gateway in the cloud, Studio local (your laptop) Run Studio on your laptop as above, then set an upstream URL your laptop can reach. Recommended (Tailscale Serve on the gateway host): 1. In Studio (on your laptop): Upstream URL: Upstream Token: your gateway token 3. Keep in mind: Studio still needs a gateway token here, even if the OpenClaw Control UI can use Tailscale identity headers Raw is an advanced/manual

path and may need extra OpenClaw origin configuration Alternative (SSH tunnel): 1. In Studio: Upstream URL: C) Studio in the cloud, Gateway in the cloud This is the “always-on” setup. When Studio and OpenClaw run on the same cloud VM, keep the OpenClaw upstream local and solve browser access to Studio separately. On the VPS that will run Studio: Run Studio (same commands as above). If OpenClaw is on that same VPS, keep Studio's upstream set to: Upstream URL

: Upstream Token: your gateway token 3. Only use a remote upstream like if Studio and OpenClaw are on different machines. Notes: Avoid serving Studio behind unless you configure and rebuild. If Studio is reachable beyond loopback, is required. If you bind Studio beyond loopback, open once from each new browser to set the Studio cookie. How It Connects (Mental Model) OpenClaw Studio now runs one runtime architecture with **two primary paths**: 1. Browser -> Studio: HTTP + SSE ( , , ) 2. Studio -> Gateway (upstream

): one server-owned WebSocket opened by the Studio Node process This is why means “gateway on the Studio host”, not “gateway on your phone”. If Studio is running on a remote machine over SSH and the terminal says , that is the remote machine. Use Tailscale Serve or an SSH tunnel to open Studio from your own laptop. Install from source (advanced) Optional setup helper in a source checkout: That writes the saved gateway URL/token for this Studio host without

opening the UI first. Configuration Paths and key settings: OpenClaw config: (or via ) Studio settings: Control-plane runtime DB: Default

Frequently asked questions

How many stars and forks are recorded?

grp06/openclaw-studio has 2,034 stars and 294 forks.

What is the repository rank?

grp06/openclaw-studio is ranked 5,204.

What language and category are recorded?

grp06/openclaw-studio has 2,034 stars, uses TypeScript and belongs to AI Engineering.

Which repository is closest in rank?

grp06/openclaw-studio is compared with nottelabs/notte, ranked 5,202 with 2,039 stars.

README

OpenClaw Studio

OpenClaw Studio is a clean web dashboard for OpenClaw. Use it to connect to your Gateway, see your agents, chat, manage approvals, and configure jobs from one place.

⭐ Drop a star to help us grow! ⭐

It helps more developers discover the project.

Get Started (Pick Your Setup)

If your Gateway is already running, pick the scenario that matches where your Gateway and Studio will run:

All setups use the same install/run path (recommended): npx -y openclaw-studio@latest

Two links matter:

  1. Browser -> Studio
  2. Studio -> Gateway

localhost always means "the Studio host." If Studio and OpenClaw share a machine, the upstream should usually stay at ws://localhost:18789 even when that machine is a cloud VM.

Requirements

  • Node.js 20.9+ (LTS recommended)
  • An OpenClaw Gateway URL + token, or a local OpenClaw install Studio can detect
  • Tailscale (optional, recommended for remote access)

A) Gateway local, Studio local (same computer)

npx -y openclaw-studio@latest
cd openclaw-studio
npm run dev
  1. Open http://localhost:3000
  2. In Studio, set:
    • Upstream URL: ws://localhost:18789
    • Upstream Token: your gateway token (for example: openclaw config get gateway.auth.token)

B) Gateway in the cloud, Studio local (your laptop)

Run Studio on your laptop as above, then set an upstream URL your laptop can reach.

Recommended (Tailscale Serve on the gateway host):

  1. On the gateway host:
    • tailscale serve --yes --bg --https 443 http://127.0.0.1:18789
  2. In Studio (on your laptop):
    • Upstream URL: wss://.ts.net
    • Upstream Token: your gateway token
  3. Keep in mind:
    • Studio still needs a gateway token here, even if the OpenClaw Control UI can use Tailscale identity headers
    • Raw ws://:18789 is an advanced/manual path and may need extra OpenClaw origin configuration

Alternative (SSH tunnel):

  1. From your laptop:
    • ssh -L 18789:127.0.0.1:18789 user@
  2. In Studio:
    • Upstream URL: ws://localhost:18789

C) Studio in the cloud, Gateway in the cloud

This is the “always-on” setup. When Studio and OpenClaw run on the same cloud VM, keep the OpenClaw upstream local and solve browser access to Studio separately.

  1. On the VPS that will run Studio:
    • Run Studio (same commands as above).
  2. If OpenClaw is on that same VPS, keep Studio's upstream set to:
    • Upstream URL: ws://localhost:18789
    • Upstream Token: your gateway token
  3. Expose Studio over tailnet HTTPS:
    • tailscale serve --yes --bg --https 443 http://127.0.0.1:3000
  4. Open Studio from your laptop/phone:
    • https://.ts.net
  5. Only use a remote upstream like wss://.ts.net if Studio and OpenClaw are on different machines.

Notes:

  • Avoid serving Studio behind /studio unless you configure basePath and rebuild.
  • If Studio is reachable beyond loopback, STUDIO_ACCESS_TOKEN is required.
  • If you bind Studio beyond loopback, open /?access_token=... once from each new browser to set the Studio cookie.

How It Connects (Mental Model)

OpenClaw Studio now runs one runtime architecture with two primary paths:

  1. Browser -> Studio: HTTP + SSE (/api/runtime/*, /api/intents/*, /api/runtime/stream)
  2. Studio -> Gateway (upstream): one server-owned WebSocket opened by the Studio Node process

This is why ws://localhost:18789 means “gateway on the Studio host”, not “gateway on your phone”.

If Studio is running on a remote machine over SSH and the terminal says Open in browser: http://localhost:3000, that localhost is the remote machine. Use Tailscale Serve or an SSH tunnel to open Studio from your own laptop.

Install from source (advanced)

git clone https://github.com/grp06/openclaw-studio.git
cd openclaw-studio
npm install
npm run dev

Optional setup helper in a source checkout:

npm run studio:setup

That writes the saved gateway URL/token for this Studio host without opening the UI first.

Configuration

Paths and key settings:

  • OpenClaw config: ~/.openclaw/openclaw.json (or via OPENCLAW_STATE_DIR)
  • Studio settings: ~/.openclaw/openclaw-studio/settings.json
  • Control-plane runtime DB: ~/.openclaw/openclaw-studio/runtime.db
  • Default gateway URL: ws://localhost:18789 (override via Studio Settings or NEXT_PUBLIC_GATEWAY_URL)
  • Domain API mode: always enabled. Studio runs on the server-owned control-plane architecture.
  • STUDIO_ACCESS_TOKEN: required when binding Studio to a public host (HOST=0.0.0.0, HOST=::, or non-loopback hostnames/IPs); optional for loopback-only binds (127.0.0.1, ::1, localhost)

Startup guard behavior:

  • npm run dev and npm run dev:turbo run verify:native-runtime:repair before server startup.
  • npm run start runs verify:native-runtime:check before startup (check-only; no dependency mutation).

Why SQLite exists now:

  • Studio’s server-owned control plane stores durable runtime projection + replay outbox in runtime.db.
  • This keeps runtime history and SSE replay deterministic across page refreshes and process restarts.

UI guide

See docs/ui-guide.md for UI workflows (agent creation, cron jobs, exec approvals).

PI + chat streaming

See docs/pi-chat-streaming.md for how Studio streams runtime events over domain SSE (/api/runtime/stream), applies replay/history, and renders tool calls, thinking traces, and final transcript lines.

Permissions + sandboxing

See docs/permissions-sandboxing.md for how agent creation choices (tool policy, sandbox config, exec approvals) flow from Studio into the OpenClaw Gateway and how upstream OpenClaw enforces them at runtime (workspaces, sandbox mounts, tool availability, and exec approval prompts).

Color system

See docs/color-system.md for the semantic color contract, status mappings, and guardrails that keep action/status/danger usage consistent across the UI.

Troubleshooting

If the UI loads but “Connect” fails, it’s usually Studio->Gateway:

  • Confirm the upstream URL/token in the UI (stored on the Studio host at /openclaw-studio/settings.json).
  • If Studio is on a remote host, remember that ws://localhost:18789 means "OpenClaw on the Studio host," not "OpenClaw on your laptop."
  • If Studio is on a remote host and you cannot open http://localhost:3000 from your laptop, expose Studio with tailscale serve --yes --bg --https 443 http://127.0.0.1:3000 or use ssh -L 3000:127.0.0.1:3000 user@host.
  • EPROTO / “wrong version number”: you used wss://... to a non-TLS endpoint (use ws://..., or put the gateway behind HTTPS).
  • .ts.net + ws://: use wss:// instead.
  • Assets 404 under /studio: serve Studio at / or configure basePath and rebuild.
  • 401 “Studio access token required”: STUDIO_ACCESS_TOKEN is enabled; open /?access_token=... once to set the cookie.
  • Helpful error codes: studio.gateway_url_missing, studio.gateway_token_missing, studio.upstream_error, studio.upstream_closed.

If startup fails with better_sqlite3.node / NODE_MODULE_VERSION mismatch:

  • Run npm run verify:native-runtime:repair
  • Confirm node and npm point at the same runtime before launching Studio:
    • node -v && node -p "process.versions.modules"
    • which node && which npm
    • If they differ (for example Homebrew npm + nvm node), run nvm use in that terminal first.
  • If it still fails, run:
    • npm rebuild better-sqlite3
    • npm install

Architecture

See ARCHITECTURE.md for details on modules and data flow.

Nearby in the ranking
Similar items