| import streamlit as st |
| import requests |
|
|
| API_URL = "http://localhost:8000" |
|
|
| st.set_page_config( |
| page_title="Agentic Coding Sandbox", |
| page_icon="β‘", |
| layout="wide", |
| ) |
|
|
| st.title("β‘ Self-Correcting Agentic Coding Sandbox") |
| st.markdown( |
| "Submit a coding task. The system generates, executes, and self-heals the code in an isolated Docker sandbox." |
| ) |
|
|
| with st.sidebar: |
| st.header("About") |
| st.markdown( |
| """ |
| **Architecture:** |
| 1. **Coder Agent** β generates Python code via LLM |
| 2. **Sandbox Executor** β runs code in isolated Docker container |
| 3. **Critic Agent** β analyzes errors and suggests fixes |
| 4. **Retry Loop** β up to 3 self-healing attempts |
| """ |
| ) |
| st.caption(f"API: {API_URL}") |
|
|
| col1, col2 = st.columns([3, 2]) |
|
|
| with col1: |
| prompt = st.text_area( |
| "Describe what you want the code to do:", |
| height=150, |
| placeholder="e.g., Plot a bar chart of the top 5 most frequent words in this text.", |
| ) |
|
|
| if st.button("Run", type="primary", disabled=not prompt): |
| with st.spinner("Generating code..."): |
| try: |
| resp = requests.post( |
| f"{API_URL}/run", |
| json={"prompt": prompt}, |
| timeout=120, |
| ) |
| resp.raise_for_status() |
| result = resp.json() |
| st.session_state["result"] = result |
| except requests.exceptions.ConnectionError: |
| st.error( |
| f"Cannot connect to API at {API_URL}. " |
| "Make sure the API server is running: `python -m src.ui.api`" |
| ) |
| except Exception as e: |
| st.error(f"Error: {e}") |
|
|
| with col2: |
| st.subheader("Result") |
|
|
| if "result" in st.session_state: |
| result = st.session_state["result"] |
| retries = result.get("retries_used", 0) |
|
|
| if result.get("success"): |
| st.success(f"β
Succeeded ({retries} retries)") |
| else: |
| st.error(f"β Failed ({retries} retries)") |
|
|
| files = result.get("files", {}) |
| if files: |
| for name, b64 in files.items(): |
| st.image(f"data:image/png;base64,{b64}", caption=name, use_container_width=True) |
|
|
| output = result.get("output") or result.get("error") or "No output" |
| if output.strip(): |
| st.code(output, language="text", line_numbers=True) |
| else: |
| st.info("Submit a prompt to see results here.") |
|
|
| st.divider() |
|
|
| st.subheader("Execution Trace") |
|
|
| if "result" in st.session_state: |
| trace = st.session_state["result"].get("trace", []) |
| if trace: |
| for i, entry in enumerate(trace): |
| node = entry.get("node", "?") |
| retry = entry.get("retry", 0) |
| icon = {"coder": "βοΈ", "executor": "βοΈ", "critic": "π"}.get(node, "β‘οΈ") |
|
|
| with st.expander(f"{icon} Attempt {retry + 1} β {node}", expanded=True): |
| st.json(entry) |
| else: |
| st.caption("No trace data available.") |
| else: |
| st.caption("Run a prompt to see the execution trace.") |
|
|