Documentation
Everything you need to run BanditDB in production — from a first predict → reward cycle to offline policy evaluation and causal validation.
Installation
Binary, Docker, Compose, Helm, or source — plus the Python SDK.
Run
Start the engine, verify health, create your first campaign.
Quick Start
A complete predict→reward cycle in a dozen lines of Python.
Errors & Status Codes
Every HTTP status the API returns, and what to do about it.
Native Agent Use (MCP)
Give any Claude-based agent persistent decision memory.
Choosing an Algorithm
LinUCB, Thompson Sampling, NeuralLinUCB, NeuralTS, or Progressive.
Model Architecture
Why Neural-Linear beats end-to-end Deep RL in production.
NeuralLinUCB
A learnable, non-linear view of the context for LinUCB.
Data Science Escape Hatch
Parquet export, offline policy evaluation, WAL inspection.
How Recovery Works
Checkpoint + WAL replay, and exactly what an acknowledged write guarantees.
Observability & Monitoring
Prometheus metrics, selection entropy, health endpoints, alerts.
Use Cases
Context + finite choices + measurable outcome = candidate.