title	Customer Service Bot — OpenEnv
emoji	🤖
colorFrom	gray
colorTo	blue
sdk	docker
app_port	8000

Customer Service Bot — OpenEnv

Run everything from the repository root:

PS C:\NPersonal\Projects\openenv-project>

Prerequisites

• Python 3.11+
• uv (recommended) or pip
• Optional: Docker for container builds
• Optional: Ollama for local LLM (OpenAI-compatible API on port 11434)

Install (root)

cd C:\NPersonal\Projects\openenv-project
uv sync

Or: python -m pip install -r requirements.txt

If a leftover empty RUNTIME\ folder exists from an older layout, close any program using it and delete the folder.

Start the API server (server/app.py)

uv run uvicorn server.app:app --host 127.0.0.1 --port 8000

Health check: http://127.0.0.1:8000/health

Baseline inference (inference.py)

Requires the server above. From the same repo root:

$env:ENV_BASE_URL = "http://127.0.0.1:8000"
$env:API_BASE_URL = "http://127.0.0.1:11434/v1"
$env:MODEL_NAME = "llama3.2"
$env:HF_TOKEN = "ollama"   # required for LLM path; no Python fallback — unset ⇒ rule-based only

uv run python inference.py

Structured logs go to stdout ([START], [STEP], [END]). Details go to stderr.

To print the full chat transcript while keeping graded stdout clean: $env:VERBOSE_CHAT = "1" before running inference.py (transcript on stderr only).

Hugging Face / remote OpenAI-compatible

Set submission variables as required by the hackathon:

• API_BASE_URL — LLM base URL
• MODEL_NAME — model id
• HF_TOKEN — API key

Docker (build from root only)

docker build -t openenv-runtime .
docker run --rm -p 8000:8000 openenv-runtime

The image uses uv to lock and install dependencies (Dockerfile).
If inference runs inside Docker but Ollama runs on the host (Windows/macOS), use e.g. API_BASE_URL=http://host.docker.internal:11434/v1 for the LLM client.

Tests

uv run python tests/test_environment.py

Layout

Path	Purpose
openenv.yaml	OpenEnv metadata
models.py	Pydantic types
server/app.py	FastAPI app (/reset, /step, /state, /health)
server/environment.py	CustomerServiceEnv
src/	Tasks, graders, simulation
inference.py	Baseline inference (root; required for submission)
Dockerfile	Production image (uv + uvicorn)

Push to Hugging Face (openenv push)

From the repo root, pass . as the directory (last argument). Example:

openenv push --repo-id YOUR_USER/YOUR_SPACE .

Log in first: huggingface-cli login (or set HF_TOKEN).

README metadata (colorFrom / colorTo)

Hugging Face only accepts certain theme colors in the YAML above (e.g. gray, red, yellow, green, blue, indigo, purple, pink). If upload fails with Invalid metadata in README.md, fix those two fields—do not use custom color names or hex.

Windows: 'charmap' codec can't encode character (emoji in output)

PowerShell’s default encoding can break openenv when it prints Unicode (e.g. icons). Use UTF-8 for that session, then push:

$env:PYTHONUTF8 = "1"
$env:PYTHONIOENCODING = "utf-8"
openenv push --repo-id YOUR_USER/YOUR_SPACE .

Or use the helper (same env vars):

.\push-to-hf.ps1 push --repo-id YOUR_USER/YOUR_SPACE .

Compliance

See COMPLIANCE.md for the pre-submission checklist (HF Space, Docker build, openenv validate, inference format, infra limits).