From 48ef57715e7b2014ef434ab2233c92f969e41cd4 Mon Sep 17 00:00:00 2001
From: dtarkent2-sys <dtarkent2@gmail.com>
Date: Fri, 20 Feb 2026 00:41:46 +0000
Subject: [PATCH] docs: add Chainlit web UI design for Railway deployment

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 .../2026-02-20-chainlit-web-ui-design.md      | 67 +++++++++++++++++++
 1 file changed, 67 insertions(+)
 create mode 100644 docs/plans/2026-02-20-chainlit-web-ui-design.md

diff --git a/docs/plans/2026-02-20-chainlit-web-ui-design.md b/docs/plans/2026-02-20-chainlit-web-ui-design.md
new file mode 100644
index 00000000..197a205e
--- /dev/null
+++ b/docs/plans/2026-02-20-chainlit-web-ui-design.md
@@ -0,0 +1,67 @@
+# TradingAgents Chainlit Web UI — Design
+
+## Summary
+
+Add a Chainlit web UI to TradingAgents so it can be deployed on Railway as a web service. Users interact via chat messages (e.g., "Analyze NVDA") and see live agent progress streamed into the browser.
+
+## Architecture
+
+Thin Chainlit wrapper around the existing `TradingAgentsGraph` programmatic API. ~150 lines of new code in a single `app.py`.
+
+## Components
+
+### `app.py` (Chainlit entry point)
+
+- `@cl.on_chat_start` — Welcome message explaining usage (e.g., "Type a ticker like `NVDA` or `Analyze AAPL 2024-12-01`")
+- `@cl.on_message` — Parse ticker + optional date from user message, create `TradingAgentsGraph` with Anthropic config, run `propagate()` in debug mode, stream Chainlit `Step` messages for each agent phase, send final decision as formatted message
+
+### `Dockerfile`
+
+- Python 3.13-slim base
+- Install requirements.txt
+- Expose `$PORT`
+- `CMD: chainlit run app.py --host 0.0.0.0 --port $PORT`
+
+### `railway.toml`
+
+- Build from Dockerfile
+- Health check on `/`
+
+### Railway Environment Variables
+
+- `ANTHROPIC_API_KEY` — required, for Claude models
+- `PORT` — auto-set by Railway
+
+## LLM Configuration
+
+- Provider: `anthropic`
+- Quick-think model: `claude-haiku-4-5-20251001`
+- Deep-think model: `claude-sonnet-4-5-20241022`
+- Data vendor: `yfinance` (no extra API keys needed)
+
+## Data Flow
+
+```
+User message: "Analyze NVDA"
+  -> Parse: ticker=NVDA, date=today
+  -> TradingAgentsGraph(config={anthropic, haiku/sonnet})
+  -> graph.propagate("NVDA", "2026-02-20")
+     -> Debug stream chunks
+     -> Each chunk -> Chainlit Step (Analyst, Research, Trading, Risk, Portfolio)
+  -> Final decision -> formatted Chainlit message with markdown
+```
+
+## Message Parsing
+
+Simple regex/string parsing:
+- `"NVDA"` -> ticker=NVDA, date=today
+- `"Analyze AAPL 2024-12-01"` -> ticker=AAPL, date=2024-12-01
+- `"What's the outlook for TSLA?"` -> ticker=TSLA, date=today
+- Extract uppercase 1-5 letter words as potential tickers
+
+## Deployment
+
+1. Push changes to `github.com/dtarkent2-sys/TradingAgents` main branch
+2. Create Railway service from GitHub repo
+3. Set `ANTHROPIC_API_KEY` env var
+4. Railway auto-deploys, Chainlit serves on assigned PORT