Back to search
glassbox
MCP serverGitHub
An AI agent you can see inside: .NET 10 agent loop, MCP tools and a live trace panel. Runs on a Jetson Nano.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/nexusframe--glassbox)HTML
<a href="/listings/nexusframe--glassbox"><img src="/api/listings/nexusframe--glassbox/badge.svg" alt="findsafeskills safety rating" /></a>Install
i
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 69 lines
- # Glassbox
- Glassbox is an AI agent you can see inside. You chat with it in the browser. Next to the chat, a live panel shows each step the agent takes: the guards, the model calls and the tool calls. The agent loop is our own code, and the tools run on a separate MCP server.
- Demo: <https://glassbox.wasyluk.pl>
- ## What the visitor sees
- ```text
- ┌─ Chat ─────────────────────┐┌─ How the agent works ───────────────┐
- │ You: Time in Tokyo, and ││ [1] question 0 ms │
- │ 17% of 240? ││ [2] L0 OK budget 12% of day │
- │ ││ [3] model ▶ 2 tools 410 ms │
- │ Agent: 23:14 in Tokyo. ││ ├─ get_datetime(Tokyo) 4 ms ✓ │
- │ 17% of 240 is 40.80. ││ └─ calculate(240*0.17) 3 ms ✓ │
- │ ││ [4] model ▶ answer 520 ms │
- │ [sample questions] [PL|EN] ││ [5] end 2 turns, 1140 tok, $0.002 │
- └────────────────────────────┘└─────────────────────────────────────┘
- ```
- ## Where it runs
- ```text
- user's browser
- │ opens page │ chat: fetch POST (stream)
- ▼ ▼
- ┌──────────────────┐ ┌───────────────────┐
- │ Cloudflare Pages │ │ Cloudflare Tunnel │ no open
- │ chat page, video │ └─────────┬─────────┘ inbound
- └──────────────────┘ ▼
- ┌─ small arm64 server (Docker) ──────┐
- │ ┌────────────┐ MCP ┌──────────┐ │
- │ │ Api + loop │ ─────▶ │ Tools │ │
- │ └─────┬──────┘ └──────────┘ │
- └───────┼────────────────────────────┘
- ▼ HTTPS
- ┌──────────────────┐
- │ OpenRouter ──▶ │
- │ cheap cloud LLM │
- └──────────────────┘
- ```
- ## Run locally
- You need Docker. Steps:
- ```sh
- cp deploy/.env.example deploy/.env # then set OPENROUTER_API_KEY
- curl localhost:8080/health
- ```
- Serve the page:
- ```sh
- python3 -m http.server 5500 -d web
- ```
- `web/config.js` points to port 8090. To use it, set `API_PORT=8090` in `deploy/.env`.
- ## Privacy
- Prompt text is not written to disk or logs.
- A public instance must name its data controller (RODO/GDPR): set `GLASSBOX_OPERATOR` and `GLASSBOX_CONTACT` (an e-mail or an `https://` URL) in `web/config.js` at deploy time. The first-view notice shows them.
- ## How it is built
- The code is written with AI coding agents; I read, review and accept every change, and I decide the direction. The ADRs in `docs/` show that work.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.