rLLM UI

Web interface for monitoring and analyzing rLLM training runs in real time. Think of wandb dedicated to rLLM, with powerful features such as episode/trajectory search, observability AI agent and more.

---

Getting Started

There are two ways to access rLLM UI:

1. Cloud — Use our hosted service at ui.rllm-project.com (see below).
2. Self-hosted — Run locally from the repository (see below).

---

Cloud Setup

1. Run rllm login
2. Sign up at ui.rllm-project.com
3. Copy your API key (shown once at registration) and paste it in terminal (or save it as RLLM_API_KEY in .env)

That's it. No need to setup the database and other configurations.

Note

The observability AI agent can be enabled by adding your ANTHROPIC_API_KEY in the Settings page in the UI — no extra configuration needed.

---

Self-hosted Setup

Quick Start (Docker)

git clone https://github.com/rllm-org/rllm-ui.git
cd rllm-ui
cp .env.example .env
# Edit .env — add ANTHROPIC_API_KEY to enable the AI agent (optional)
docker-compose up

Open http://localhost:3000. Data is stored in a Docker volume and persists across restarts.

With BigQuery Traces

If your agent traces are in BigQuery, use the BigQuery overlay to mount your GCP credentials:

# Authenticate with GCP (one-time)
gcloud auth application-default login

# Start with BigQuery support
docker-compose -f docker-compose.yml -f docker-compose.bigquery.yml up

Then configure your BQ project, dataset, and table in the UI — see the Agent Observability Guide for detailed steps.

Alternatively, for service account key auth, see the comments in docker-compose.bigquery.yml.

Development without Docker

# Install dependencies
cd api && pip install -r requirements.txt
cd ../frontend && npm install

# Run (two terminals)
cd api && uvicorn main:app --reload --port 8000
cd frontend && npm run dev

Open http://localhost:5173.

Database

• SQLite (default) — No setup required. Created automatically on first run.
• PostgreSQL — Adds full-text search with stemming and relevance ranking. Set DATABASE_URL in .env:

DATABASE_URL="postgresql://user:pass@localhost:5432/rllm_ui"

Configuration

Variable	Required	Default	Description
ANTHROPIC_API_KEY	No	—	Enables the built-in observability AI agent
DATABASE_URL	No	SQLite	PostgreSQL connection string
RLLM_UI_URL	No	http://localhost:3000	URL of your rllm-ui server (for training scripts)

BigQuery project, dataset, and table are configured in the UI (Settings page or on first use). See the Agent Observability Guide.

---

Connecting rLLM to UI

Training runs with script

Regardless of the service (cloud or self-hosted) you use, add ui to your trainer's logger list in your rLLM training script:

trainer.logger="['console','wandb','ui']"

Training / Evaluation runs with rLLM CLI

If using our cloud service and rLLM CLI, you can run training and eval runs as such:

rllm train [dataset name]
rllm eval [dataset name]

If logged in, traces will automatically stream to the UI.

---

How It Works

rLLM connects to the UI via the UILogger backend, registered as "ui" in the Tracking class (rllm/utils/tracking.py).

On init, the logger:

1. Creates a training session via POST /api/sessions
2. Starts a background heartbeat thread (for crash detection)
3. Wraps stdout/stderr with TeeStream to capture training logs

During training, the logger sends data over HTTP.

So the overall flow looks like:

---